git.oblomov.eu Git - linux-2.6/blob - drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2008 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include <linux/module.h>
  30 #include <linux/types.h>
  31 #include <linux/init.h>
  32 #include <linux/pci.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/delay.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/tcp.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/mii.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <linux/cpu.h>
  45 #include <linux/smp.h>
  46 #include <linux/pm_qos_params.h>
  47
  48 #include "e1000.h"
  49
  50 #define DRV_VERSION "0.3.3.3-k6"
  51 char e1000e_driver_name[] = "e1000e";
  52 const char e1000e_driver_version[] = DRV_VERSION;
  53
  54 static const struct e1000_info *e1000_info_tbl[] = {
  55         [board_82571]           = &e1000_82571_info,
  56         [board_82572]           = &e1000_82572_info,
  57         [board_82573]           = &e1000_82573_info,
  58         [board_82574]           = &e1000_82574_info,
  59         [board_80003es2lan]     = &e1000_es2_info,
  60         [board_ich8lan]         = &e1000_ich8_info,
  61         [board_ich9lan]         = &e1000_ich9_info,
  62         [board_ich10lan]        = &e1000_ich10_info,
  63 };
  64
  65 #ifdef DEBUG
  66 /**
  67  * e1000_get_hw_dev_name - return device name string
  68  * used by hardware layer to print debugging information
  69  **/
  70 char *e1000e_get_hw_dev_name(struct e1000_hw *hw)
  71 {
  72         return hw->adapter->netdev->name;
  73 }
  74 #endif
  75
  76 /**
  77  * e1000_desc_unused - calculate if we have unused descriptors
  78  **/
  79 static int e1000_desc_unused(struct e1000_ring *ring)
  80 {
  81         if (ring->next_to_clean > ring->next_to_use)
  82                 return ring->next_to_clean - ring->next_to_use - 1;
  83
  84         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
  85 }
  86
  87 /**
  88  * e1000_receive_skb - helper function to handle Rx indications
  89  * @adapter: board private structure
  90  * @status: descriptor status field as written by hardware
  91  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
  92  * @skb: pointer to sk_buff to be indicated to stack
  93  **/
  94 static void e1000_receive_skb(struct e1000_adapter *adapter,
  95                               struct net_device *netdev,
  96                               struct sk_buff *skb,
  97                               u8 status, __le16 vlan)
  98 {
  99         skb->protocol = eth_type_trans(skb, netdev);
 100
 101         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 102                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
 103                                          le16_to_cpu(vlan));
 104         else
 105                 netif_receive_skb(skb);
 106 }
 107
 108 /**
 109  * e1000_rx_checksum - Receive Checksum Offload for 82543
 110  * @adapter:     board private structure
 111  * @status_err:  receive descriptor status and error fields
 112  * @csum:       receive descriptor csum field
 113  * @sk_buff:     socket buffer with received data
 114  **/
 115 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 116                               u32 csum, struct sk_buff *skb)
 117 {
 118         u16 status = (u16)status_err;
 119         u8 errors = (u8)(status_err >> 24);
 120         skb->ip_summed = CHECKSUM_NONE;
 121
 122         /* Ignore Checksum bit is set */
 123         if (status & E1000_RXD_STAT_IXSM)
 124                 return;
 125         /* TCP/UDP checksum error bit is set */
 126         if (errors & E1000_RXD_ERR_TCPE) {
 127                 /* let the stack verify checksum errors */
 128                 adapter->hw_csum_err++;
 129                 return;
 130         }
 131
 132         /* TCP/UDP Checksum has not been calculated */
 133         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 134                 return;
 135
 136         /* It must be a TCP or UDP packet with a valid checksum */
 137         if (status & E1000_RXD_STAT_TCPCS) {
 138                 /* TCP checksum is good */
 139                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 140         } else {
 141                 /*
 142                  * IP fragment with UDP payload
 143                  * Hardware complements the payload checksum, so we undo it
 144                  * and then put the value in host order for further stack use.
 145                  */
 146                 __sum16 sum = (__force __sum16)htons(csum);
 147                 skb->csum = csum_unfold(~sum);
 148                 skb->ip_summed = CHECKSUM_COMPLETE;
 149         }
 150         adapter->hw_csum_good++;
 151 }
 152
 153 /**
 154  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 155  * @adapter: address of board private structure
 156  **/
 157 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 158                                    int cleaned_count)
 159 {
 160         struct net_device *netdev = adapter->netdev;
 161         struct pci_dev *pdev = adapter->pdev;
 162         struct e1000_ring *rx_ring = adapter->rx_ring;
 163         struct e1000_rx_desc *rx_desc;
 164         struct e1000_buffer *buffer_info;
 165         struct sk_buff *skb;
 166         unsigned int i;
 167         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
 168
 169         i = rx_ring->next_to_use;
 170         buffer_info = &rx_ring->buffer_info[i];
 171
 172         while (cleaned_count--) {
 173                 skb = buffer_info->skb;
 174                 if (skb) {
 175                         skb_trim(skb, 0);
 176                         goto map_skb;
 177                 }
 178
 179                 skb = netdev_alloc_skb(netdev, bufsz);
 180                 if (!skb) {
 181                         /* Better luck next round */
 182                         adapter->alloc_rx_buff_failed++;
 183                         break;
 184                 }
 185
 186                 /*
 187                  * Make buffer alignment 2 beyond a 16 byte boundary
 188                  * this will result in a 16 byte aligned IP header after
 189                  * the 14 byte MAC header is removed
 190                  */
 191                 skb_reserve(skb, NET_IP_ALIGN);
 192
 193                 buffer_info->skb = skb;
 194 map_skb:
 195                 buffer_info->dma = pci_map_single(pdev, skb->data,
 196                                                   adapter->rx_buffer_len,
 197                                                   PCI_DMA_FROMDEVICE);
 198                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 199                         dev_err(&pdev->dev, "RX DMA map failed\n");
 200                         adapter->rx_dma_failed++;
 201                         break;
 202                 }
 203
 204                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 205                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 206
 207                 i++;
 208                 if (i == rx_ring->count)
 209                         i = 0;
 210                 buffer_info = &rx_ring->buffer_info[i];
 211         }
 212
 213         if (rx_ring->next_to_use != i) {
 214                 rx_ring->next_to_use = i;
 215                 if (i-- == 0)
 216                         i = (rx_ring->count - 1);
 217
 218                 /*
 219                  * Force memory writes to complete before letting h/w
 220                  * know there are new descriptors to fetch.  (Only
 221                  * applicable for weak-ordered memory model archs,
 222                  * such as IA-64).
 223                  */
 224                 wmb();
 225                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 226         }
 227 }
 228
 229 /**
 230  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 231  * @adapter: address of board private structure
 232  **/
 233 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 234                                       int cleaned_count)
 235 {
 236         struct net_device *netdev = adapter->netdev;
 237         struct pci_dev *pdev = adapter->pdev;
 238         union e1000_rx_desc_packet_split *rx_desc;
 239         struct e1000_ring *rx_ring = adapter->rx_ring;
 240         struct e1000_buffer *buffer_info;
 241         struct e1000_ps_page *ps_page;
 242         struct sk_buff *skb;
 243         unsigned int i, j;
 244
 245         i = rx_ring->next_to_use;
 246         buffer_info = &rx_ring->buffer_info[i];
 247
 248         while (cleaned_count--) {
 249                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 250
 251                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 252                         ps_page = &buffer_info->ps_pages[j];
 253                         if (j >= adapter->rx_ps_pages) {
 254                                 /* all unused desc entries get hw null ptr */
 255                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 256                                 continue;
 257                         }
 258                         if (!ps_page->page) {
 259                                 ps_page->page = alloc_page(GFP_ATOMIC);
 260                                 if (!ps_page->page) {
 261                                         adapter->alloc_rx_buff_failed++;
 262                                         goto no_buffers;
 263                                 }
 264                                 ps_page->dma = pci_map_page(pdev,
 265                                                    ps_page->page,
 266                                                    0, PAGE_SIZE,
 267                                                    PCI_DMA_FROMDEVICE);
 268                                 if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 269                                         dev_err(&adapter->pdev->dev,
 270                                           "RX DMA page map failed\n");
 271                                         adapter->rx_dma_failed++;
 272                                         goto no_buffers;
 273                                 }
 274                         }
 275                         /*
 276                          * Refresh the desc even if buffer_addrs
 277                          * didn't change because each write-back
 278                          * erases this info.
 279                          */
 280                         rx_desc->read.buffer_addr[j+1] =
 281                              cpu_to_le64(ps_page->dma);
 282                 }
 283
 284                 skb = netdev_alloc_skb(netdev,
 285                                        adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 286
 287                 if (!skb) {
 288                         adapter->alloc_rx_buff_failed++;
 289                         break;
 290                 }
 291
 292                 /*
 293                  * Make buffer alignment 2 beyond a 16 byte boundary
 294                  * this will result in a 16 byte aligned IP header after
 295                  * the 14 byte MAC header is removed
 296                  */
 297                 skb_reserve(skb, NET_IP_ALIGN);
 298
 299                 buffer_info->skb = skb;
 300                 buffer_info->dma = pci_map_single(pdev, skb->data,
 301                                                   adapter->rx_ps_bsize0,
 302                                                   PCI_DMA_FROMDEVICE);
 303                 if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 304                         dev_err(&pdev->dev, "RX DMA map failed\n");
 305                         adapter->rx_dma_failed++;
 306                         /* cleanup skb */
 307                         dev_kfree_skb_any(skb);
 308                         buffer_info->skb = NULL;
 309                         break;
 310                 }
 311
 312                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 313
 314                 i++;
 315                 if (i == rx_ring->count)
 316                         i = 0;
 317                 buffer_info = &rx_ring->buffer_info[i];
 318         }
 319
 320 no_buffers:
 321         if (rx_ring->next_to_use != i) {
 322                 rx_ring->next_to_use = i;
 323
 324                 if (!(i--))
 325                         i = (rx_ring->count - 1);
 326
 327                 /*
 328                  * Force memory writes to complete before letting h/w
 329                  * know there are new descriptors to fetch.  (Only
 330                  * applicable for weak-ordered memory model archs,
 331                  * such as IA-64).
 332                  */
 333                 wmb();
 334                 /*
 335                  * Hardware increments by 16 bytes, but packet split
 336                  * descriptors are 32 bytes...so we increment tail
 337                  * twice as much.
 338                  */
 339                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 340         }
 341 }
 342
 343 /**
 344  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 345  * @adapter: address of board private structure
 346  * @rx_ring: pointer to receive ring structure
 347  * @cleaned_count: number of buffers to allocate this pass
 348  **/
 349
 350 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 351                                          int cleaned_count)
 352 {
 353         struct net_device *netdev = adapter->netdev;
 354         struct pci_dev *pdev = adapter->pdev;
 355         struct e1000_rx_desc *rx_desc;
 356         struct e1000_ring *rx_ring = adapter->rx_ring;
 357         struct e1000_buffer *buffer_info;
 358         struct sk_buff *skb;
 359         unsigned int i;
 360         unsigned int bufsz = 256 -
 361                              16 /* for skb_reserve */ -
 362                              NET_IP_ALIGN;
 363
 364         i = rx_ring->next_to_use;
 365         buffer_info = &rx_ring->buffer_info[i];
 366
 367         while (cleaned_count--) {
 368                 skb = buffer_info->skb;
 369                 if (skb) {
 370                         skb_trim(skb, 0);
 371                         goto check_page;
 372                 }
 373
 374                 skb = netdev_alloc_skb(netdev, bufsz);
 375                 if (unlikely(!skb)) {
 376                         /* Better luck next round */
 377                         adapter->alloc_rx_buff_failed++;
 378                         break;
 379                 }
 380
 381                 /* Make buffer alignment 2 beyond a 16 byte boundary
 382                  * this will result in a 16 byte aligned IP header after
 383                  * the 14 byte MAC header is removed
 384                  */
 385                 skb_reserve(skb, NET_IP_ALIGN);
 386
 387                 buffer_info->skb = skb;
 388 check_page:
 389                 /* allocate a new page if necessary */
 390                 if (!buffer_info->page) {
 391                         buffer_info->page = alloc_page(GFP_ATOMIC);
 392                         if (unlikely(!buffer_info->page)) {
 393                                 adapter->alloc_rx_buff_failed++;
 394                                 break;
 395                         }
 396                 }
 397
 398                 if (!buffer_info->dma)
 399                         buffer_info->dma = pci_map_page(pdev,
 400                                                         buffer_info->page, 0,
 401                                                         PAGE_SIZE,
 402                                                         PCI_DMA_FROMDEVICE);
 403
 404                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 405                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 406
 407                 if (unlikely(++i == rx_ring->count))
 408                         i = 0;
 409                 buffer_info = &rx_ring->buffer_info[i];
 410         }
 411
 412         if (likely(rx_ring->next_to_use != i)) {
 413                 rx_ring->next_to_use = i;
 414                 if (unlikely(i-- == 0))
 415                         i = (rx_ring->count - 1);
 416
 417                 /* Force memory writes to complete before letting h/w
 418                  * know there are new descriptors to fetch.  (Only
 419                  * applicable for weak-ordered memory model archs,
 420                  * such as IA-64). */
 421                 wmb();
 422                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 423         }
 424 }
 425
 426 /**
 427  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 428  * @adapter: board private structure
 429  *
 430  * the return value indicates whether actual cleaning was done, there
 431  * is no guarantee that everything was cleaned
 432  **/
 433 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 434                                int *work_done, int work_to_do)
 435 {
 436         struct net_device *netdev = adapter->netdev;
 437         struct pci_dev *pdev = adapter->pdev;
 438         struct e1000_ring *rx_ring = adapter->rx_ring;
 439         struct e1000_rx_desc *rx_desc, *next_rxd;
 440         struct e1000_buffer *buffer_info, *next_buffer;
 441         u32 length;
 442         unsigned int i;
 443         int cleaned_count = 0;
 444         bool cleaned = 0;
 445         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 446
 447         i = rx_ring->next_to_clean;
 448         rx_desc = E1000_RX_DESC(*rx_ring, i);
 449         buffer_info = &rx_ring->buffer_info[i];
 450
 451         while (rx_desc->status & E1000_RXD_STAT_DD) {
 452                 struct sk_buff *skb;
 453                 u8 status;
 454
 455                 if (*work_done >= work_to_do)
 456                         break;
 457                 (*work_done)++;
 458
 459                 status = rx_desc->status;
 460                 skb = buffer_info->skb;
 461                 buffer_info->skb = NULL;
 462
 463                 prefetch(skb->data - NET_IP_ALIGN);
 464
 465                 i++;
 466                 if (i == rx_ring->count)
 467                         i = 0;
 468                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 469                 prefetch(next_rxd);
 470
 471                 next_buffer = &rx_ring->buffer_info[i];
 472
 473                 cleaned = 1;
 474                 cleaned_count++;
 475                 pci_unmap_single(pdev,
 476                                  buffer_info->dma,
 477                                  adapter->rx_buffer_len,
 478                                  PCI_DMA_FROMDEVICE);
 479                 buffer_info->dma = 0;
 480
 481                 length = le16_to_cpu(rx_desc->length);
 482
 483                 /* !EOP means multiple descriptors were used to store a single
 484                  * packet, also make sure the frame isn't just CRC only */
 485                 if (!(status & E1000_RXD_STAT_EOP) || (length <= 4)) {
 486                         /* All receives must fit into a single buffer */
 487                         e_dbg("%s: Receive packet consumed multiple buffers\n",
 488                               netdev->name);
 489                         /* recycle */
 490                         buffer_info->skb = skb;
 491                         goto next_desc;
 492                 }
 493
 494                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 495                         /* recycle */
 496                         buffer_info->skb = skb;
 497                         goto next_desc;
 498                 }
 499
 500                 total_rx_bytes += length;
 501                 total_rx_packets++;
 502
 503                 /*
 504                  * code added for copybreak, this should improve
 505                  * performance for small packets with large amounts
 506                  * of reassembly being done in the stack
 507                  */
 508                 if (length < copybreak) {
 509                         struct sk_buff *new_skb =
 510                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
 511                         if (new_skb) {
 512                                 skb_reserve(new_skb, NET_IP_ALIGN);
 513                                 skb_copy_to_linear_data_offset(new_skb,
 514                                                                -NET_IP_ALIGN,
 515                                                                (skb->data -
 516                                                                 NET_IP_ALIGN),
 517                                                                (length +
 518                                                                 NET_IP_ALIGN));
 519                                 /* save the skb in buffer_info as good */
 520                                 buffer_info->skb = skb;
 521                                 skb = new_skb;
 522                         }
 523                         /* else just continue with the old one */
 524                 }
 525                 /* end copybreak code */
 526                 skb_put(skb, length);
 527
 528                 /* Receive Checksum Offload */
 529                 e1000_rx_checksum(adapter,
 530                                   (u32)(status) |
 531                                   ((u32)(rx_desc->errors) << 24),
 532                                   le16_to_cpu(rx_desc->csum), skb);
 533
 534                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 535
 536 next_desc:
 537                 rx_desc->status = 0;
 538
 539                 /* return some buffers to hardware, one at a time is too slow */
 540                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 541                         adapter->alloc_rx_buf(adapter, cleaned_count);
 542                         cleaned_count = 0;
 543                 }
 544
 545                 /* use prefetched values */
 546                 rx_desc = next_rxd;
 547                 buffer_info = next_buffer;
 548         }
 549         rx_ring->next_to_clean = i;
 550
 551         cleaned_count = e1000_desc_unused(rx_ring);
 552         if (cleaned_count)
 553                 adapter->alloc_rx_buf(adapter, cleaned_count);
 554
 555         adapter->total_rx_bytes += total_rx_bytes;
 556         adapter->total_rx_packets += total_rx_packets;
 557         adapter->net_stats.rx_bytes += total_rx_bytes;
 558         adapter->net_stats.rx_packets += total_rx_packets;
 559         return cleaned;
 560 }
 561
 562 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 563                              struct e1000_buffer *buffer_info)
 564 {
 565         if (buffer_info->dma) {
 566                 pci_unmap_page(adapter->pdev, buffer_info->dma,
 567                                buffer_info->length, PCI_DMA_TODEVICE);
 568                 buffer_info->dma = 0;
 569         }
 570         if (buffer_info->skb) {
 571                 dev_kfree_skb_any(buffer_info->skb);
 572                 buffer_info->skb = NULL;
 573         }
 574 }
 575
 576 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
 577 {
 578         struct e1000_ring *tx_ring = adapter->tx_ring;
 579         unsigned int i = tx_ring->next_to_clean;
 580         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 581         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 582
 583         /* detected Tx unit hang */
 584         e_err("Detected Tx Unit Hang:\n"
 585               "  TDH                  <%x>\n"
 586               "  TDT                  <%x>\n"
 587               "  next_to_use          <%x>\n"
 588               "  next_to_clean        <%x>\n"
 589               "buffer_info[next_to_clean]:\n"
 590               "  time_stamp           <%lx>\n"
 591               "  next_to_watch        <%x>\n"
 592               "  jiffies              <%lx>\n"
 593               "  next_to_watch.status <%x>\n",
 594               readl(adapter->hw.hw_addr + tx_ring->head),
 595               readl(adapter->hw.hw_addr + tx_ring->tail),
 596               tx_ring->next_to_use,
 597               tx_ring->next_to_clean,
 598               tx_ring->buffer_info[eop].time_stamp,
 599               eop,
 600               jiffies,
 601               eop_desc->upper.fields.status);
 602 }
 603
 604 /**
 605  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 606  * @adapter: board private structure
 607  *
 608  * the return value indicates whether actual cleaning was done, there
 609  * is no guarantee that everything was cleaned
 610  **/
 611 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 612 {
 613         struct net_device *netdev = adapter->netdev;
 614         struct e1000_hw *hw = &adapter->hw;
 615         struct e1000_ring *tx_ring = adapter->tx_ring;
 616         struct e1000_tx_desc *tx_desc, *eop_desc;
 617         struct e1000_buffer *buffer_info;
 618         unsigned int i, eop;
 619         unsigned int count = 0;
 620         bool cleaned = 0;
 621         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 622
 623         i = tx_ring->next_to_clean;
 624         eop = tx_ring->buffer_info[i].next_to_watch;
 625         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 626
 627         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
 628                 for (cleaned = 0; !cleaned; ) {
 629                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 630                         buffer_info = &tx_ring->buffer_info[i];
 631                         cleaned = (i == eop);
 632
 633                         if (cleaned) {
 634                                 struct sk_buff *skb = buffer_info->skb;
 635                                 unsigned int segs, bytecount;
 636                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 637                                 /* multiply data chunks by size of headers */
 638                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 639                                             skb->len;
 640                                 total_tx_packets += segs;
 641                                 total_tx_bytes += bytecount;
 642                         }
 643
 644                         e1000_put_txbuf(adapter, buffer_info);
 645                         tx_desc->upper.data = 0;
 646
 647                         i++;
 648                         if (i == tx_ring->count)
 649                                 i = 0;
 650                 }
 651
 652                 eop = tx_ring->buffer_info[i].next_to_watch;
 653                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
 654 #define E1000_TX_WEIGHT 64
 655                 /* weight of a sort for tx, to avoid endless transmit cleanup */
 656                 if (count++ == E1000_TX_WEIGHT)
 657                         break;
 658         }
 659
 660         tx_ring->next_to_clean = i;
 661
 662 #define TX_WAKE_THRESHOLD 32
 663         if (cleaned && netif_carrier_ok(netdev) &&
 664                      e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
 665                 /* Make sure that anybody stopping the queue after this
 666                  * sees the new next_to_clean.
 667                  */
 668                 smp_mb();
 669
 670                 if (netif_queue_stopped(netdev) &&
 671                     !(test_bit(__E1000_DOWN, &adapter->state))) {
 672                         netif_wake_queue(netdev);
 673                         ++adapter->restart_queue;
 674                 }
 675         }
 676
 677         if (adapter->detect_tx_hung) {
 678                 /*
 679                  * Detect a transmit hang in hardware, this serializes the
 680                  * check with the clearing of time_stamp and movement of i
 681                  */
 682                 adapter->detect_tx_hung = 0;
 683                 if (tx_ring->buffer_info[eop].dma &&
 684                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
 685                                + (adapter->tx_timeout_factor * HZ))
 686                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 687                         e1000_print_tx_hang(adapter);
 688                         netif_stop_queue(netdev);
 689                 }
 690         }
 691         adapter->total_tx_bytes += total_tx_bytes;
 692         adapter->total_tx_packets += total_tx_packets;
 693         adapter->net_stats.tx_bytes += total_tx_bytes;
 694         adapter->net_stats.tx_packets += total_tx_packets;
 695         return cleaned;
 696 }
 697
 698 /**
 699  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
 700  * @adapter: board private structure
 701  *
 702  * the return value indicates whether actual cleaning was done, there
 703  * is no guarantee that everything was cleaned
 704  **/
 705 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 706                                   int *work_done, int work_to_do)
 707 {
 708         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
 709         struct net_device *netdev = adapter->netdev;
 710         struct pci_dev *pdev = adapter->pdev;
 711         struct e1000_ring *rx_ring = adapter->rx_ring;
 712         struct e1000_buffer *buffer_info, *next_buffer;
 713         struct e1000_ps_page *ps_page;
 714         struct sk_buff *skb;
 715         unsigned int i, j;
 716         u32 length, staterr;
 717         int cleaned_count = 0;
 718         bool cleaned = 0;
 719         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 720
 721         i = rx_ring->next_to_clean;
 722         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 723         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 724         buffer_info = &rx_ring->buffer_info[i];
 725
 726         while (staterr & E1000_RXD_STAT_DD) {
 727                 if (*work_done >= work_to_do)
 728                         break;
 729                 (*work_done)++;
 730                 skb = buffer_info->skb;
 731
 732                 /* in the packet split case this is header only */
 733                 prefetch(skb->data - NET_IP_ALIGN);
 734
 735                 i++;
 736                 if (i == rx_ring->count)
 737                         i = 0;
 738                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
 739                 prefetch(next_rxd);
 740
 741                 next_buffer = &rx_ring->buffer_info[i];
 742
 743                 cleaned = 1;
 744                 cleaned_count++;
 745                 pci_unmap_single(pdev, buffer_info->dma,
 746                                  adapter->rx_ps_bsize0,
 747                                  PCI_DMA_FROMDEVICE);
 748                 buffer_info->dma = 0;
 749
 750                 if (!(staterr & E1000_RXD_STAT_EOP)) {
 751                         e_dbg("%s: Packet Split buffers didn't pick up the "
 752                               "full packet\n", netdev->name);
 753                         dev_kfree_skb_irq(skb);
 754                         goto next_desc;
 755                 }
 756
 757                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 758                         dev_kfree_skb_irq(skb);
 759                         goto next_desc;
 760                 }
 761
 762                 length = le16_to_cpu(rx_desc->wb.middle.length0);
 763
 764                 if (!length) {
 765                         e_dbg("%s: Last part of the packet spanning multiple "
 766                               "descriptors\n", netdev->name);
 767                         dev_kfree_skb_irq(skb);
 768                         goto next_desc;
 769                 }
 770
 771                 /* Good Receive */
 772                 skb_put(skb, length);
 773
 774                 {
 775                 /*
 776                  * this looks ugly, but it seems compiler issues make it
 777                  * more efficient than reusing j
 778                  */
 779                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
 780
 781                 /*
 782                  * page alloc/put takes too long and effects small packet
 783                  * throughput, so unsplit small packets and save the alloc/put
 784                  * only valid in softirq (napi) context to call kmap_*
 785                  */
 786                 if (l1 && (l1 <= copybreak) &&
 787                     ((length + l1) <= adapter->rx_ps_bsize0)) {
 788                         u8 *vaddr;
 789
 790                         ps_page = &buffer_info->ps_pages[0];
 791
 792                         /*
 793                          * there is no documentation about how to call
 794                          * kmap_atomic, so we can't hold the mapping
 795                          * very long
 796                          */
 797                         pci_dma_sync_single_for_cpu(pdev, ps_page->dma,
 798                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 799                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
 800                         memcpy(skb_tail_pointer(skb), vaddr, l1);
 801                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
 802                         pci_dma_sync_single_for_device(pdev, ps_page->dma,
 803                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
 804
 805                         skb_put(skb, l1);
 806                         goto copydone;
 807                 } /* if */
 808                 }
 809
 810                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 811                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
 812                         if (!length)
 813                                 break;
 814
 815                         ps_page = &buffer_info->ps_pages[j];
 816                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 817                                        PCI_DMA_FROMDEVICE);
 818                         ps_page->dma = 0;
 819                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 820                         ps_page->page = NULL;
 821                         skb->len += length;
 822                         skb->data_len += length;
 823                         skb->truesize += length;
 824                 }
 825
 826 copydone:
 827                 total_rx_bytes += skb->len;
 828                 total_rx_packets++;
 829
 830                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
 831                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
 832
 833                 if (rx_desc->wb.upper.header_status &
 834                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
 835                         adapter->rx_hdr_split++;
 836
 837                 e1000_receive_skb(adapter, netdev, skb,
 838                                   staterr, rx_desc->wb.middle.vlan);
 839
 840 next_desc:
 841                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
 842                 buffer_info->skb = NULL;
 843
 844                 /* return some buffers to hardware, one at a time is too slow */
 845                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 846                         adapter->alloc_rx_buf(adapter, cleaned_count);
 847                         cleaned_count = 0;
 848                 }
 849
 850                 /* use prefetched values */
 851                 rx_desc = next_rxd;
 852                 buffer_info = next_buffer;
 853
 854                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
 855         }
 856         rx_ring->next_to_clean = i;
 857
 858         cleaned_count = e1000_desc_unused(rx_ring);
 859         if (cleaned_count)
 860                 adapter->alloc_rx_buf(adapter, cleaned_count);
 861
 862         adapter->total_rx_bytes += total_rx_bytes;
 863         adapter->total_rx_packets += total_rx_packets;
 864         adapter->net_stats.rx_bytes += total_rx_bytes;
 865         adapter->net_stats.rx_packets += total_rx_packets;
 866         return cleaned;
 867 }
 868
 869 /**
 870  * e1000_consume_page - helper function
 871  **/
 872 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
 873                                u16 length)
 874 {
 875         bi->page = NULL;
 876         skb->len += length;
 877         skb->data_len += length;
 878         skb->truesize += length;
 879 }
 880
 881 /**
 882  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 883  * @adapter: board private structure
 884  *
 885  * the return value indicates whether actual cleaning was done, there
 886  * is no guarantee that everything was cleaned
 887  **/
 888
 889 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 890                                      int *work_done, int work_to_do)
 891 {
 892         struct net_device *netdev = adapter->netdev;
 893         struct pci_dev *pdev = adapter->pdev;
 894         struct e1000_ring *rx_ring = adapter->rx_ring;
 895         struct e1000_rx_desc *rx_desc, *next_rxd;
 896         struct e1000_buffer *buffer_info, *next_buffer;
 897         u32 length;
 898         unsigned int i;
 899         int cleaned_count = 0;
 900         bool cleaned = false;
 901         unsigned int total_rx_bytes=0, total_rx_packets=0;
 902
 903         i = rx_ring->next_to_clean;
 904         rx_desc = E1000_RX_DESC(*rx_ring, i);
 905         buffer_info = &rx_ring->buffer_info[i];
 906
 907         while (rx_desc->status & E1000_RXD_STAT_DD) {
 908                 struct sk_buff *skb;
 909                 u8 status;
 910
 911                 if (*work_done >= work_to_do)
 912                         break;
 913                 (*work_done)++;
 914
 915                 status = rx_desc->status;
 916                 skb = buffer_info->skb;
 917                 buffer_info->skb = NULL;
 918
 919                 ++i;
 920                 if (i == rx_ring->count)
 921                         i = 0;
 922                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 923                 prefetch(next_rxd);
 924
 925                 next_buffer = &rx_ring->buffer_info[i];
 926
 927                 cleaned = true;
 928                 cleaned_count++;
 929                 pci_unmap_page(pdev, buffer_info->dma, PAGE_SIZE,
 930                                PCI_DMA_FROMDEVICE);
 931                 buffer_info->dma = 0;
 932
 933                 length = le16_to_cpu(rx_desc->length);
 934
 935                 /* errors is only valid for DD + EOP descriptors */
 936                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
 937                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
 938                                 /* recycle both page and skb */
 939                                 buffer_info->skb = skb;
 940                                 /* an error means any chain goes out the window
 941                                  * too */
 942                                 if (rx_ring->rx_skb_top)
 943                                         dev_kfree_skb(rx_ring->rx_skb_top);
 944                                 rx_ring->rx_skb_top = NULL;
 945                                 goto next_desc;
 946                 }
 947
 948 #define rxtop rx_ring->rx_skb_top
 949                 if (!(status & E1000_RXD_STAT_EOP)) {
 950                         /* this descriptor is only the beginning (or middle) */
 951                         if (!rxtop) {
 952                                 /* this is the beginning of a chain */
 953                                 rxtop = skb;
 954                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
 955                                                    0, length);
 956                         } else {
 957                                 /* this is the middle of a chain */
 958                                 skb_fill_page_desc(rxtop,
 959                                     skb_shinfo(rxtop)->nr_frags,
 960                                     buffer_info->page, 0, length);
 961                                 /* re-use the skb, only consumed the page */
 962                                 buffer_info->skb = skb;
 963                         }
 964                         e1000_consume_page(buffer_info, rxtop, length);
 965                         goto next_desc;
 966                 } else {
 967                         if (rxtop) {
 968                                 /* end of the chain */
 969                                 skb_fill_page_desc(rxtop,
 970                                     skb_shinfo(rxtop)->nr_frags,
 971                                     buffer_info->page, 0, length);
 972                                 /* re-use the current skb, we only consumed the
 973                                  * page */
 974                                 buffer_info->skb = skb;
 975                                 skb = rxtop;
 976                                 rxtop = NULL;
 977                                 e1000_consume_page(buffer_info, skb, length);
 978                         } else {
 979                                 /* no chain, got EOP, this buf is the packet
 980                                  * copybreak to save the put_page/alloc_page */
 981                                 if (length <= copybreak &&
 982                                     skb_tailroom(skb) >= length) {
 983                                         u8 *vaddr;
 984                                         vaddr = kmap_atomic(buffer_info->page,
 985                                                            KM_SKB_DATA_SOFTIRQ);
 986                                         memcpy(skb_tail_pointer(skb), vaddr,
 987                                                length);
 988                                         kunmap_atomic(vaddr,
 989                                                       KM_SKB_DATA_SOFTIRQ);
 990                                         /* re-use the page, so don't erase
 991                                          * buffer_info->page */
 992                                         skb_put(skb, length);
 993                                 } else {
 994                                         skb_fill_page_desc(skb, 0,
 995                                                            buffer_info->page, 0,
 996                                                            length);
 997                                         e1000_consume_page(buffer_info, skb,
 998                                                            length);
 999                                 }
1000                         }
1001                 }
1002
1003                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1004                 e1000_rx_checksum(adapter,
1005                                   (u32)(status) |
1006                                   ((u32)(rx_desc->errors) << 24),
1007                                   le16_to_cpu(rx_desc->csum), skb);
1008
1009                 /* probably a little skewed due to removing CRC */
1010                 total_rx_bytes += skb->len;
1011                 total_rx_packets++;
1012
1013                 /* eth type trans needs skb->data to point to something */
1014                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1015                         e_err("pskb_may_pull failed.\n");
1016                         dev_kfree_skb(skb);
1017                         goto next_desc;
1018                 }
1019
1020                 e1000_receive_skb(adapter, netdev, skb, status,
1021                                   rx_desc->special);
1022
1023 next_desc:
1024                 rx_desc->status = 0;
1025
1026                 /* return some buffers to hardware, one at a time is too slow */
1027                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1028                         adapter->alloc_rx_buf(adapter, cleaned_count);
1029                         cleaned_count = 0;
1030                 }
1031
1032                 /* use prefetched values */
1033                 rx_desc = next_rxd;
1034                 buffer_info = next_buffer;
1035         }
1036         rx_ring->next_to_clean = i;
1037
1038         cleaned_count = e1000_desc_unused(rx_ring);
1039         if (cleaned_count)
1040                 adapter->alloc_rx_buf(adapter, cleaned_count);
1041
1042         adapter->total_rx_bytes += total_rx_bytes;
1043         adapter->total_rx_packets += total_rx_packets;
1044         adapter->net_stats.rx_bytes += total_rx_bytes;
1045         adapter->net_stats.rx_packets += total_rx_packets;
1046         return cleaned;
1047 }
1048
1049 /**
1050  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1051  * @adapter: board private structure
1052  **/
1053 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1054 {
1055         struct e1000_ring *rx_ring = adapter->rx_ring;
1056         struct e1000_buffer *buffer_info;
1057         struct e1000_ps_page *ps_page;
1058         struct pci_dev *pdev = adapter->pdev;
1059         unsigned int i, j;
1060
1061         /* Free all the Rx ring sk_buffs */
1062         for (i = 0; i < rx_ring->count; i++) {
1063                 buffer_info = &rx_ring->buffer_info[i];
1064                 if (buffer_info->dma) {
1065                         if (adapter->clean_rx == e1000_clean_rx_irq)
1066                                 pci_unmap_single(pdev, buffer_info->dma,
1067                                                  adapter->rx_buffer_len,
1068                                                  PCI_DMA_FROMDEVICE);
1069                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1070                                 pci_unmap_page(pdev, buffer_info->dma,
1071                                                PAGE_SIZE,
1072                                                PCI_DMA_FROMDEVICE);
1073                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1074                                 pci_unmap_single(pdev, buffer_info->dma,
1075                                                  adapter->rx_ps_bsize0,
1076                                                  PCI_DMA_FROMDEVICE);
1077                         buffer_info->dma = 0;
1078                 }
1079
1080                 if (buffer_info->page) {
1081                         put_page(buffer_info->page);
1082                         buffer_info->page = NULL;
1083                 }
1084
1085                 if (buffer_info->skb) {
1086                         dev_kfree_skb(buffer_info->skb);
1087                         buffer_info->skb = NULL;
1088                 }
1089
1090                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1091                         ps_page = &buffer_info->ps_pages[j];
1092                         if (!ps_page->page)
1093                                 break;
1094                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
1095                                        PCI_DMA_FROMDEVICE);
1096                         ps_page->dma = 0;
1097                         put_page(ps_page->page);
1098                         ps_page->page = NULL;
1099                 }
1100         }
1101
1102         /* there also may be some cached data from a chained receive */
1103         if (rx_ring->rx_skb_top) {
1104                 dev_kfree_skb(rx_ring->rx_skb_top);
1105                 rx_ring->rx_skb_top = NULL;
1106         }
1107
1108         /* Zero out the descriptor ring */
1109         memset(rx_ring->desc, 0, rx_ring->size);
1110
1111         rx_ring->next_to_clean = 0;
1112         rx_ring->next_to_use = 0;
1113
1114         writel(0, adapter->hw.hw_addr + rx_ring->head);
1115         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1116 }
1117
1118 static void e1000e_downshift_workaround(struct work_struct *work)
1119 {
1120         struct e1000_adapter *adapter = container_of(work,
1121                                         struct e1000_adapter, downshift_task);
1122
1123         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1124 }
1125
1126 /**
1127  * e1000_intr_msi - Interrupt Handler
1128  * @irq: interrupt number
1129  * @data: pointer to a network interface device structure
1130  **/
1131 static irqreturn_t e1000_intr_msi(int irq, void *data)
1132 {
1133         struct net_device *netdev = data;
1134         struct e1000_adapter *adapter = netdev_priv(netdev);
1135         struct e1000_hw *hw = &adapter->hw;
1136         u32 icr = er32(ICR);
1137
1138         /*
1139          * read ICR disables interrupts using IAM
1140          */
1141
1142         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1143                 hw->mac.get_link_status = 1;
1144                 /*
1145                  * ICH8 workaround-- Call gig speed drop workaround on cable
1146                  * disconnect (LSC) before accessing any PHY registers
1147                  */
1148                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1149                     (!(er32(STATUS) & E1000_STATUS_LU)))
1150                         schedule_work(&adapter->downshift_task);
1151
1152                 /*
1153                  * 80003ES2LAN workaround-- For packet buffer work-around on
1154                  * link down event; disable receives here in the ISR and reset
1155                  * adapter in watchdog
1156                  */
1157                 if (netif_carrier_ok(netdev) &&
1158                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1159                         /* disable receives */
1160                         u32 rctl = er32(RCTL);
1161                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1162                         adapter->flags |= FLAG_RX_RESTART_NOW;
1163                 }
1164                 /* guard against interrupt when we're going down */
1165                 if (!test_bit(__E1000_DOWN, &adapter->state))
1166                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1167         }
1168
1169         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1170                 adapter->total_tx_bytes = 0;
1171                 adapter->total_tx_packets = 0;
1172                 adapter->total_rx_bytes = 0;
1173                 adapter->total_rx_packets = 0;
1174                 __netif_rx_schedule(netdev, &adapter->napi);
1175         }
1176
1177         return IRQ_HANDLED;
1178 }
1179
1180 /**
1181  * e1000_intr - Interrupt Handler
1182  * @irq: interrupt number
1183  * @data: pointer to a network interface device structure
1184  **/
1185 static irqreturn_t e1000_intr(int irq, void *data)
1186 {
1187         struct net_device *netdev = data;
1188         struct e1000_adapter *adapter = netdev_priv(netdev);
1189         struct e1000_hw *hw = &adapter->hw;
1190         u32 rctl, icr = er32(ICR);
1191
1192         if (!icr)
1193                 return IRQ_NONE;  /* Not our interrupt */
1194
1195         /*
1196          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1197          * not set, then the adapter didn't send an interrupt
1198          */
1199         if (!(icr & E1000_ICR_INT_ASSERTED))
1200                 return IRQ_NONE;
1201
1202         /*
1203          * Interrupt Auto-Mask...upon reading ICR,
1204          * interrupts are masked.  No need for the
1205          * IMC write
1206          */
1207
1208         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1209                 hw->mac.get_link_status = 1;
1210                 /*
1211                  * ICH8 workaround-- Call gig speed drop workaround on cable
1212                  * disconnect (LSC) before accessing any PHY registers
1213                  */
1214                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1215                     (!(er32(STATUS) & E1000_STATUS_LU)))
1216                         schedule_work(&adapter->downshift_task);
1217
1218                 /*
1219                  * 80003ES2LAN workaround--
1220                  * For packet buffer work-around on link down event;
1221                  * disable receives here in the ISR and
1222                  * reset adapter in watchdog
1223                  */
1224                 if (netif_carrier_ok(netdev) &&
1225                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1226                         /* disable receives */
1227                         rctl = er32(RCTL);
1228                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1229                         adapter->flags |= FLAG_RX_RESTART_NOW;
1230                 }
1231                 /* guard against interrupt when we're going down */
1232                 if (!test_bit(__E1000_DOWN, &adapter->state))
1233                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1234         }
1235
1236         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1237                 adapter->total_tx_bytes = 0;
1238                 adapter->total_tx_packets = 0;
1239                 adapter->total_rx_bytes = 0;
1240                 adapter->total_rx_packets = 0;
1241                 __netif_rx_schedule(netdev, &adapter->napi);
1242         }
1243
1244         return IRQ_HANDLED;
1245 }
1246
1247 static irqreturn_t e1000_msix_other(int irq, void *data)
1248 {
1249         struct net_device *netdev = data;
1250         struct e1000_adapter *adapter = netdev_priv(netdev);
1251         struct e1000_hw *hw = &adapter->hw;
1252         u32 icr = er32(ICR);
1253
1254         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1255                 ew32(IMS, E1000_IMS_OTHER);
1256                 return IRQ_NONE;
1257         }
1258
1259         if (icr & adapter->eiac_mask)
1260                 ew32(ICS, (icr & adapter->eiac_mask));
1261
1262         if (icr & E1000_ICR_OTHER) {
1263                 if (!(icr & E1000_ICR_LSC))
1264                         goto no_link_interrupt;
1265                 hw->mac.get_link_status = 1;
1266                 /* guard against interrupt when we're going down */
1267                 if (!test_bit(__E1000_DOWN, &adapter->state))
1268                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1269         }
1270
1271 no_link_interrupt:
1272         ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1273
1274         return IRQ_HANDLED;
1275 }
1276
1277
1278 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1279 {
1280         struct net_device *netdev = data;
1281         struct e1000_adapter *adapter = netdev_priv(netdev);
1282         struct e1000_hw *hw = &adapter->hw;
1283         struct e1000_ring *tx_ring = adapter->tx_ring;
1284
1285
1286         adapter->total_tx_bytes = 0;
1287         adapter->total_tx_packets = 0;
1288
1289         if (!e1000_clean_tx_irq(adapter))
1290                 /* Ring was not completely cleaned, so fire another interrupt */
1291                 ew32(ICS, tx_ring->ims_val);
1292
1293         return IRQ_HANDLED;
1294 }
1295
1296 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1297 {
1298         struct net_device *netdev = data;
1299         struct e1000_adapter *adapter = netdev_priv(netdev);
1300
1301         /* Write the ITR value calculated at the end of the
1302          * previous interrupt.
1303          */
1304         if (adapter->rx_ring->set_itr) {
1305                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1306                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1307                 adapter->rx_ring->set_itr = 0;
1308         }
1309
1310         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1311                 adapter->total_rx_bytes = 0;
1312                 adapter->total_rx_packets = 0;
1313                 __netif_rx_schedule(netdev, &adapter->napi);
1314         }
1315         return IRQ_HANDLED;
1316 }
1317
1318 /**
1319  * e1000_configure_msix - Configure MSI-X hardware
1320  *
1321  * e1000_configure_msix sets up the hardware to properly
1322  * generate MSI-X interrupts.
1323  **/
1324 static void e1000_configure_msix(struct e1000_adapter *adapter)
1325 {
1326         struct e1000_hw *hw = &adapter->hw;
1327         struct e1000_ring *rx_ring = adapter->rx_ring;
1328         struct e1000_ring *tx_ring = adapter->tx_ring;
1329         int vector = 0;
1330         u32 ctrl_ext, ivar = 0;
1331
1332         adapter->eiac_mask = 0;
1333
1334         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1335         if (hw->mac.type == e1000_82574) {
1336                 u32 rfctl = er32(RFCTL);
1337                 rfctl |= E1000_RFCTL_ACK_DIS;
1338                 ew32(RFCTL, rfctl);
1339         }
1340
1341 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1342         /* Configure Rx vector */
1343         rx_ring->ims_val = E1000_IMS_RXQ0;
1344         adapter->eiac_mask |= rx_ring->ims_val;
1345         if (rx_ring->itr_val)
1346                 writel(1000000000 / (rx_ring->itr_val * 256),
1347                        hw->hw_addr + rx_ring->itr_register);
1348         else
1349                 writel(1, hw->hw_addr + rx_ring->itr_register);
1350         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1351
1352         /* Configure Tx vector */
1353         tx_ring->ims_val = E1000_IMS_TXQ0;
1354         vector++;
1355         if (tx_ring->itr_val)
1356                 writel(1000000000 / (tx_ring->itr_val * 256),
1357                        hw->hw_addr + tx_ring->itr_register);
1358         else
1359                 writel(1, hw->hw_addr + tx_ring->itr_register);
1360         adapter->eiac_mask |= tx_ring->ims_val;
1361         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1362
1363         /* set vector for Other Causes, e.g. link changes */
1364         vector++;
1365         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1366         if (rx_ring->itr_val)
1367                 writel(1000000000 / (rx_ring->itr_val * 256),
1368                        hw->hw_addr + E1000_EITR_82574(vector));
1369         else
1370                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1371
1372         /* Cause Tx interrupts on every write back */
1373         ivar |= (1 << 31);
1374
1375         ew32(IVAR, ivar);
1376
1377         /* enable MSI-X PBA support */
1378         ctrl_ext = er32(CTRL_EXT);
1379         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1380
1381         /* Auto-Mask Other interrupts upon ICR read */
1382 #define E1000_EIAC_MASK_82574   0x01F00000
1383         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1384         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1385         ew32(CTRL_EXT, ctrl_ext);
1386         e1e_flush();
1387 }
1388
1389 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1390 {
1391         if (adapter->msix_entries) {
1392                 pci_disable_msix(adapter->pdev);
1393                 kfree(adapter->msix_entries);
1394                 adapter->msix_entries = NULL;
1395         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1396                 pci_disable_msi(adapter->pdev);
1397                 adapter->flags &= ~FLAG_MSI_ENABLED;
1398         }
1399
1400         return;
1401 }
1402
1403 /**
1404  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1405  *
1406  * Attempt to configure interrupts using the best available
1407  * capabilities of the hardware and kernel.
1408  **/
1409 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1410 {
1411         int err;
1412         int numvecs, i;
1413
1414
1415         switch (adapter->int_mode) {
1416         case E1000E_INT_MODE_MSIX:
1417                 if (adapter->flags & FLAG_HAS_MSIX) {
1418                         numvecs = 3; /* RxQ0, TxQ0 and other */
1419                         adapter->msix_entries = kcalloc(numvecs,
1420                                                       sizeof(struct msix_entry),
1421                                                       GFP_KERNEL);
1422                         if (adapter->msix_entries) {
1423                                 for (i = 0; i < numvecs; i++)
1424                                         adapter->msix_entries[i].entry = i;
1425
1426                                 err = pci_enable_msix(adapter->pdev,
1427                                                       adapter->msix_entries,
1428                                                       numvecs);
1429                                 if (err == 0)
1430                                         return;
1431                         }
1432                         /* MSI-X failed, so fall through and try MSI */
1433                         e_err("Failed to initialize MSI-X interrupts.  "
1434                               "Falling back to MSI interrupts.\n");
1435                         e1000e_reset_interrupt_capability(adapter);
1436                 }
1437                 adapter->int_mode = E1000E_INT_MODE_MSI;
1438                 /* Fall through */
1439         case E1000E_INT_MODE_MSI:
1440                 if (!pci_enable_msi(adapter->pdev)) {
1441                         adapter->flags |= FLAG_MSI_ENABLED;
1442                 } else {
1443                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1444                         e_err("Failed to initialize MSI interrupts.  Falling "
1445                               "back to legacy interrupts.\n");
1446                 }
1447                 /* Fall through */
1448         case E1000E_INT_MODE_LEGACY:
1449                 /* Don't do anything; this is the system default */
1450                 break;
1451         }
1452
1453         return;
1454 }
1455
1456 /**
1457  * e1000_request_msix - Initialize MSI-X interrupts
1458  *
1459  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1460  * kernel.
1461  **/
1462 static int e1000_request_msix(struct e1000_adapter *adapter)
1463 {
1464         struct net_device *netdev = adapter->netdev;
1465         int err = 0, vector = 0;
1466
1467         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1468                 sprintf(adapter->rx_ring->name, "%s-rx0", netdev->name);
1469         else
1470                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1471         err = request_irq(adapter->msix_entries[vector].vector,
1472                           &e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1473                           netdev);
1474         if (err)
1475                 goto out;
1476         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1477         adapter->rx_ring->itr_val = adapter->itr;
1478         vector++;
1479
1480         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1481                 sprintf(adapter->tx_ring->name, "%s-tx0", netdev->name);
1482         else
1483                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1484         err = request_irq(adapter->msix_entries[vector].vector,
1485                           &e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1486                           netdev);
1487         if (err)
1488                 goto out;
1489         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1490         adapter->tx_ring->itr_val = adapter->itr;
1491         vector++;
1492
1493         err = request_irq(adapter->msix_entries[vector].vector,
1494                           &e1000_msix_other, 0, netdev->name, netdev);
1495         if (err)
1496                 goto out;
1497
1498         e1000_configure_msix(adapter);
1499         return 0;
1500 out:
1501         return err;
1502 }
1503
1504 /**
1505  * e1000_request_irq - initialize interrupts
1506  *
1507  * Attempts to configure interrupts using the best available
1508  * capabilities of the hardware and kernel.
1509  **/
1510 static int e1000_request_irq(struct e1000_adapter *adapter)
1511 {
1512         struct net_device *netdev = adapter->netdev;
1513         int err;
1514
1515         if (adapter->msix_entries) {
1516                 err = e1000_request_msix(adapter);
1517                 if (!err)
1518                         return err;
1519                 /* fall back to MSI */
1520                 e1000e_reset_interrupt_capability(adapter);
1521                 adapter->int_mode = E1000E_INT_MODE_MSI;
1522                 e1000e_set_interrupt_capability(adapter);
1523         }
1524         if (adapter->flags & FLAG_MSI_ENABLED) {
1525                 err = request_irq(adapter->pdev->irq, &e1000_intr_msi, 0,
1526                                   netdev->name, netdev);
1527                 if (!err)
1528                         return err;
1529
1530                 /* fall back to legacy interrupt */
1531                 e1000e_reset_interrupt_capability(adapter);
1532                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1533         }
1534
1535         err = request_irq(adapter->pdev->irq, &e1000_intr, IRQF_SHARED,
1536                           netdev->name, netdev);
1537         if (err)
1538                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1539
1540         return err;
1541 }
1542
1543 static void e1000_free_irq(struct e1000_adapter *adapter)
1544 {
1545         struct net_device *netdev = adapter->netdev;
1546
1547         if (adapter->msix_entries) {
1548                 int vector = 0;
1549
1550                 free_irq(adapter->msix_entries[vector].vector, netdev);
1551                 vector++;
1552
1553                 free_irq(adapter->msix_entries[vector].vector, netdev);
1554                 vector++;
1555
1556                 /* Other Causes interrupt vector */
1557                 free_irq(adapter->msix_entries[vector].vector, netdev);
1558                 return;
1559         }
1560
1561         free_irq(adapter->pdev->irq, netdev);
1562 }
1563
1564 /**
1565  * e1000_irq_disable - Mask off interrupt generation on the NIC
1566  **/
1567 static void e1000_irq_disable(struct e1000_adapter *adapter)
1568 {
1569         struct e1000_hw *hw = &adapter->hw;
1570
1571         ew32(IMC, ~0);
1572         if (adapter->msix_entries)
1573                 ew32(EIAC_82574, 0);
1574         e1e_flush();
1575         synchronize_irq(adapter->pdev->irq);
1576 }
1577
1578 /**
1579  * e1000_irq_enable - Enable default interrupt generation settings
1580  **/
1581 static void e1000_irq_enable(struct e1000_adapter *adapter)
1582 {
1583         struct e1000_hw *hw = &adapter->hw;
1584
1585         if (adapter->msix_entries) {
1586                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1587                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1588         } else {
1589                 ew32(IMS, IMS_ENABLE_MASK);
1590         }
1591         e1e_flush();
1592 }
1593
1594 /**
1595  * e1000_get_hw_control - get control of the h/w from f/w
1596  * @adapter: address of board private structure
1597  *
1598  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1599  * For ASF and Pass Through versions of f/w this means that
1600  * the driver is loaded. For AMT version (only with 82573)
1601  * of the f/w this means that the network i/f is open.
1602  **/
1603 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1604 {
1605         struct e1000_hw *hw = &adapter->hw;
1606         u32 ctrl_ext;
1607         u32 swsm;
1608
1609         /* Let firmware know the driver has taken over */
1610         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1611                 swsm = er32(SWSM);
1612                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1613         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1614                 ctrl_ext = er32(CTRL_EXT);
1615                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1616         }
1617 }
1618
1619 /**
1620  * e1000_release_hw_control - release control of the h/w to f/w
1621  * @adapter: address of board private structure
1622  *
1623  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1624  * For ASF and Pass Through versions of f/w this means that the
1625  * driver is no longer loaded. For AMT version (only with 82573) i
1626  * of the f/w this means that the network i/f is closed.
1627  *
1628  **/
1629 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1630 {
1631         struct e1000_hw *hw = &adapter->hw;
1632         u32 ctrl_ext;
1633         u32 swsm;
1634
1635         /* Let firmware taken over control of h/w */
1636         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1637                 swsm = er32(SWSM);
1638                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1639         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1640                 ctrl_ext = er32(CTRL_EXT);
1641                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1642         }
1643 }
1644
1645 /**
1646  * @e1000_alloc_ring - allocate memory for a ring structure
1647  **/
1648 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
1649                                 struct e1000_ring *ring)
1650 {
1651         struct pci_dev *pdev = adapter->pdev;
1652
1653         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
1654                                         GFP_KERNEL);
1655         if (!ring->desc)
1656                 return -ENOMEM;
1657
1658         return 0;
1659 }
1660
1661 /**
1662  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
1663  * @adapter: board private structure
1664  *
1665  * Return 0 on success, negative on failure
1666  **/
1667 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
1668 {
1669         struct e1000_ring *tx_ring = adapter->tx_ring;
1670         int err = -ENOMEM, size;
1671
1672         size = sizeof(struct e1000_buffer) * tx_ring->count;
1673         tx_ring->buffer_info = vmalloc(size);
1674         if (!tx_ring->buffer_info)
1675                 goto err;
1676         memset(tx_ring->buffer_info, 0, size);
1677
1678         /* round up to nearest 4K */
1679         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1680         tx_ring->size = ALIGN(tx_ring->size, 4096);
1681
1682         err = e1000_alloc_ring_dma(adapter, tx_ring);
1683         if (err)
1684                 goto err;
1685
1686         tx_ring->next_to_use = 0;
1687         tx_ring->next_to_clean = 0;
1688         spin_lock_init(&adapter->tx_queue_lock);
1689
1690         return 0;
1691 err:
1692         vfree(tx_ring->buffer_info);
1693         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1694         return err;
1695 }
1696
1697 /**
1698  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
1699  * @adapter: board private structure
1700  *
1701  * Returns 0 on success, negative on failure
1702  **/
1703 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
1704 {
1705         struct e1000_ring *rx_ring = adapter->rx_ring;
1706         struct e1000_buffer *buffer_info;
1707         int i, size, desc_len, err = -ENOMEM;
1708
1709         size = sizeof(struct e1000_buffer) * rx_ring->count;
1710         rx_ring->buffer_info = vmalloc(size);
1711         if (!rx_ring->buffer_info)
1712                 goto err;
1713         memset(rx_ring->buffer_info, 0, size);
1714
1715         for (i = 0; i < rx_ring->count; i++) {
1716                 buffer_info = &rx_ring->buffer_info[i];
1717                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
1718                                                 sizeof(struct e1000_ps_page),
1719                                                 GFP_KERNEL);
1720                 if (!buffer_info->ps_pages)
1721                         goto err_pages;
1722         }
1723
1724         desc_len = sizeof(union e1000_rx_desc_packet_split);
1725
1726         /* Round up to nearest 4K */
1727         rx_ring->size = rx_ring->count * desc_len;
1728         rx_ring->size = ALIGN(rx_ring->size, 4096);
1729
1730         err = e1000_alloc_ring_dma(adapter, rx_ring);
1731         if (err)
1732                 goto err_pages;
1733
1734         rx_ring->next_to_clean = 0;
1735         rx_ring->next_to_use = 0;
1736         rx_ring->rx_skb_top = NULL;
1737
1738         return 0;
1739
1740 err_pages:
1741         for (i = 0; i < rx_ring->count; i++) {
1742                 buffer_info = &rx_ring->buffer_info[i];
1743                 kfree(buffer_info->ps_pages);
1744         }
1745 err:
1746         vfree(rx_ring->buffer_info);
1747         e_err("Unable to allocate memory for the transmit descriptor ring\n");
1748         return err;
1749 }
1750
1751 /**
1752  * e1000_clean_tx_ring - Free Tx Buffers
1753  * @adapter: board private structure
1754  **/
1755 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
1756 {
1757         struct e1000_ring *tx_ring = adapter->tx_ring;
1758         struct e1000_buffer *buffer_info;
1759         unsigned long size;
1760         unsigned int i;
1761
1762         for (i = 0; i < tx_ring->count; i++) {
1763                 buffer_info = &tx_ring->buffer_info[i];
1764                 e1000_put_txbuf(adapter, buffer_info);
1765         }
1766
1767         size = sizeof(struct e1000_buffer) * tx_ring->count;
1768         memset(tx_ring->buffer_info, 0, size);
1769
1770         memset(tx_ring->desc, 0, tx_ring->size);
1771
1772         tx_ring->next_to_use = 0;
1773         tx_ring->next_to_clean = 0;
1774
1775         writel(0, adapter->hw.hw_addr + tx_ring->head);
1776         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1777 }
1778
1779 /**
1780  * e1000e_free_tx_resources - Free Tx Resources per Queue
1781  * @adapter: board private structure
1782  *
1783  * Free all transmit software resources
1784  **/
1785 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
1786 {
1787         struct pci_dev *pdev = adapter->pdev;
1788         struct e1000_ring *tx_ring = adapter->tx_ring;
1789
1790         e1000_clean_tx_ring(adapter);
1791
1792         vfree(tx_ring->buffer_info);
1793         tx_ring->buffer_info = NULL;
1794
1795         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1796                           tx_ring->dma);
1797         tx_ring->desc = NULL;
1798 }
1799
1800 /**
1801  * e1000e_free_rx_resources - Free Rx Resources
1802  * @adapter: board private structure
1803  *
1804  * Free all receive software resources
1805  **/
1806
1807 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
1808 {
1809         struct pci_dev *pdev = adapter->pdev;
1810         struct e1000_ring *rx_ring = adapter->rx_ring;
1811         int i;
1812
1813         e1000_clean_rx_ring(adapter);
1814
1815         for (i = 0; i < rx_ring->count; i++) {
1816                 kfree(rx_ring->buffer_info[i].ps_pages);
1817         }
1818
1819         vfree(rx_ring->buffer_info);
1820         rx_ring->buffer_info = NULL;
1821
1822         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1823                           rx_ring->dma);
1824         rx_ring->desc = NULL;
1825 }
1826
1827 /**
1828  * e1000_update_itr - update the dynamic ITR value based on statistics
1829  * @adapter: pointer to adapter
1830  * @itr_setting: current adapter->itr
1831  * @packets: the number of packets during this measurement interval
1832  * @bytes: the number of bytes during this measurement interval
1833  *
1834  *      Stores a new ITR value based on packets and byte
1835  *      counts during the last interrupt.  The advantage of per interrupt
1836  *      computation is faster updates and more accurate ITR for the current
1837  *      traffic pattern.  Constants in this function were computed
1838  *      based on theoretical maximum wire speed and thresholds were set based
1839  *      on testing data as well as attempting to minimize response time
1840  *      while increasing bulk throughput.  This functionality is controlled
1841  *      by the InterruptThrottleRate module parameter.
1842  **/
1843 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
1844                                      u16 itr_setting, int packets,
1845                                      int bytes)
1846 {
1847         unsigned int retval = itr_setting;
1848
1849         if (packets == 0)
1850                 goto update_itr_done;
1851
1852         switch (itr_setting) {
1853         case lowest_latency:
1854                 /* handle TSO and jumbo frames */
1855                 if (bytes/packets > 8000)
1856                         retval = bulk_latency;
1857                 else if ((packets < 5) && (bytes > 512)) {
1858                         retval = low_latency;
1859                 }
1860                 break;
1861         case low_latency:  /* 50 usec aka 20000 ints/s */
1862                 if (bytes > 10000) {
1863                         /* this if handles the TSO accounting */
1864                         if (bytes/packets > 8000) {
1865                                 retval = bulk_latency;
1866                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
1867                                 retval = bulk_latency;
1868                         } else if ((packets > 35)) {
1869                                 retval = lowest_latency;
1870                         }
1871                 } else if (bytes/packets > 2000) {
1872                         retval = bulk_latency;
1873                 } else if (packets <= 2 && bytes < 512) {
1874                         retval = lowest_latency;
1875                 }
1876                 break;
1877         case bulk_latency: /* 250 usec aka 4000 ints/s */
1878                 if (bytes > 25000) {
1879                         if (packets > 35) {
1880                                 retval = low_latency;
1881                         }
1882                 } else if (bytes < 6000) {
1883                         retval = low_latency;
1884                 }
1885                 break;
1886         }
1887
1888 update_itr_done:
1889         return retval;
1890 }
1891
1892 static void e1000_set_itr(struct e1000_adapter *adapter)
1893 {
1894         struct e1000_hw *hw = &adapter->hw;
1895         u16 current_itr;
1896         u32 new_itr = adapter->itr;
1897
1898         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
1899         if (adapter->link_speed != SPEED_1000) {
1900                 current_itr = 0;
1901                 new_itr = 4000;
1902                 goto set_itr_now;
1903         }
1904
1905         adapter->tx_itr = e1000_update_itr(adapter,
1906                                     adapter->tx_itr,
1907                                     adapter->total_tx_packets,
1908                                     adapter->total_tx_bytes);
1909         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1910         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
1911                 adapter->tx_itr = low_latency;
1912
1913         adapter->rx_itr = e1000_update_itr(adapter,
1914                                     adapter->rx_itr,
1915                                     adapter->total_rx_packets,
1916                                     adapter->total_rx_bytes);
1917         /* conservative mode (itr 3) eliminates the lowest_latency setting */
1918         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
1919                 adapter->rx_itr = low_latency;
1920
1921         current_itr = max(adapter->rx_itr, adapter->tx_itr);
1922
1923         switch (current_itr) {
1924         /* counts and packets in update_itr are dependent on these numbers */
1925         case lowest_latency:
1926                 new_itr = 70000;
1927                 break;
1928         case low_latency:
1929                 new_itr = 20000; /* aka hwitr = ~200 */
1930                 break;
1931         case bulk_latency:
1932                 new_itr = 4000;
1933                 break;
1934         default:
1935                 break;
1936         }
1937
1938 set_itr_now:
1939         if (new_itr != adapter->itr) {
1940                 /*
1941                  * this attempts to bias the interrupt rate towards Bulk
1942                  * by adding intermediate steps when interrupt rate is
1943                  * increasing
1944                  */
1945                 new_itr = new_itr > adapter->itr ?
1946                              min(adapter->itr + (new_itr >> 2), new_itr) :
1947                              new_itr;
1948                 adapter->itr = new_itr;
1949                 adapter->rx_ring->itr_val = new_itr;
1950                 if (adapter->msix_entries)
1951                         adapter->rx_ring->set_itr = 1;
1952                 else
1953                         ew32(ITR, 1000000000 / (new_itr * 256));
1954         }
1955 }
1956
1957 /**
1958  * e1000_alloc_queues - Allocate memory for all rings
1959  * @adapter: board private structure to initialize
1960  **/
1961 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1962 {
1963         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1964         if (!adapter->tx_ring)
1965                 goto err;
1966
1967         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
1968         if (!adapter->rx_ring)
1969                 goto err;
1970
1971         return 0;
1972 err:
1973         e_err("Unable to allocate memory for queues\n");
1974         kfree(adapter->rx_ring);
1975         kfree(adapter->tx_ring);
1976         return -ENOMEM;
1977 }
1978
1979 /**
1980  * e1000_clean - NAPI Rx polling callback
1981  * @napi: struct associated with this polling callback
1982  * @budget: amount of packets driver is allowed to process this poll
1983  **/
1984 static int e1000_clean(struct napi_struct *napi, int budget)
1985 {
1986         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
1987         struct e1000_hw *hw = &adapter->hw;
1988         struct net_device *poll_dev = adapter->netdev;
1989         int tx_cleaned = 0, work_done = 0;
1990
1991         /* Must NOT use netdev_priv macro here. */
1992         adapter = poll_dev->priv;
1993
1994         if (adapter->msix_entries &&
1995             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
1996                 goto clean_rx;
1997
1998         /*
1999          * e1000_clean is called per-cpu.  This lock protects
2000          * tx_ring from being cleaned by multiple cpus
2001          * simultaneously.  A failure obtaining the lock means
2002          * tx_ring is currently being cleaned anyway.
2003          */
2004         if (spin_trylock(&adapter->tx_queue_lock)) {
2005                 tx_cleaned = e1000_clean_tx_irq(adapter);
2006                 spin_unlock(&adapter->tx_queue_lock);
2007         }
2008
2009 clean_rx:
2010         adapter->clean_rx(adapter, &work_done, budget);
2011
2012         if (tx_cleaned)
2013                 work_done = budget;
2014
2015         /* If budget not fully consumed, exit the polling mode */
2016         if (work_done < budget) {
2017                 if (adapter->itr_setting & 3)
2018                         e1000_set_itr(adapter);
2019                 netif_rx_complete(poll_dev, napi);
2020                 if (adapter->msix_entries)
2021                         ew32(IMS, adapter->rx_ring->ims_val);
2022                 else
2023                         e1000_irq_enable(adapter);
2024         }
2025
2026         return work_done;
2027 }
2028
2029 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2030 {
2031         struct e1000_adapter *adapter = netdev_priv(netdev);
2032         struct e1000_hw *hw = &adapter->hw;
2033         u32 vfta, index;
2034
2035         /* don't update vlan cookie if already programmed */
2036         if ((adapter->hw.mng_cookie.status &
2037              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2038             (vid == adapter->mng_vlan_id))
2039                 return;
2040         /* add VID to filter table */
2041         index = (vid >> 5) & 0x7F;
2042         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2043         vfta |= (1 << (vid & 0x1F));
2044         e1000e_write_vfta(hw, index, vfta);
2045 }
2046
2047 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2048 {
2049         struct e1000_adapter *adapter = netdev_priv(netdev);
2050         struct e1000_hw *hw = &adapter->hw;
2051         u32 vfta, index;
2052
2053         if (!test_bit(__E1000_DOWN, &adapter->state))
2054                 e1000_irq_disable(adapter);
2055         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2056
2057         if (!test_bit(__E1000_DOWN, &adapter->state))
2058                 e1000_irq_enable(adapter);
2059
2060         if ((adapter->hw.mng_cookie.status &
2061              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2062             (vid == adapter->mng_vlan_id)) {
2063                 /* release control to f/w */
2064                 e1000_release_hw_control(adapter);
2065                 return;
2066         }
2067
2068         /* remove VID from filter table */
2069         index = (vid >> 5) & 0x7F;
2070         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2071         vfta &= ~(1 << (vid & 0x1F));
2072         e1000e_write_vfta(hw, index, vfta);
2073 }
2074
2075 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2076 {
2077         struct net_device *netdev = adapter->netdev;
2078         u16 vid = adapter->hw.mng_cookie.vlan_id;
2079         u16 old_vid = adapter->mng_vlan_id;
2080
2081         if (!adapter->vlgrp)
2082                 return;
2083
2084         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2085                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2086                 if (adapter->hw.mng_cookie.status &
2087                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2088                         e1000_vlan_rx_add_vid(netdev, vid);
2089                         adapter->mng_vlan_id = vid;
2090                 }
2091
2092                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2093                                 (vid != old_vid) &&
2094                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2095                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2096         } else {
2097                 adapter->mng_vlan_id = vid;
2098         }
2099 }
2100
2101
2102 static void e1000_vlan_rx_register(struct net_device *netdev,
2103                                    struct vlan_group *grp)
2104 {
2105         struct e1000_adapter *adapter = netdev_priv(netdev);
2106         struct e1000_hw *hw = &adapter->hw;
2107         u32 ctrl, rctl;
2108
2109         if (!test_bit(__E1000_DOWN, &adapter->state))
2110                 e1000_irq_disable(adapter);
2111         adapter->vlgrp = grp;
2112
2113         if (grp) {
2114                 /* enable VLAN tag insert/strip */
2115                 ctrl = er32(CTRL);
2116                 ctrl |= E1000_CTRL_VME;
2117                 ew32(CTRL, ctrl);
2118
2119                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2120                         /* enable VLAN receive filtering */
2121                         rctl = er32(RCTL);
2122                         rctl &= ~E1000_RCTL_CFIEN;
2123                         ew32(RCTL, rctl);
2124                         e1000_update_mng_vlan(adapter);
2125                 }
2126         } else {
2127                 /* disable VLAN tag insert/strip */
2128                 ctrl = er32(CTRL);
2129                 ctrl &= ~E1000_CTRL_VME;
2130                 ew32(CTRL, ctrl);
2131
2132                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2133                         if (adapter->mng_vlan_id !=
2134                             (u16)E1000_MNG_VLAN_NONE) {
2135                                 e1000_vlan_rx_kill_vid(netdev,
2136                                                        adapter->mng_vlan_id);
2137                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2138                         }
2139                 }
2140         }
2141
2142         if (!test_bit(__E1000_DOWN, &adapter->state))
2143                 e1000_irq_enable(adapter);
2144 }
2145
2146 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2147 {
2148         u16 vid;
2149
2150         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2151
2152         if (!adapter->vlgrp)
2153                 return;
2154
2155         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2156                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2157                         continue;
2158                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2159         }
2160 }
2161
2162 static void e1000_init_manageability(struct e1000_adapter *adapter)
2163 {
2164         struct e1000_hw *hw = &adapter->hw;
2165         u32 manc, manc2h;
2166
2167         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2168                 return;
2169
2170         manc = er32(MANC);
2171
2172         /*
2173          * enable receiving management packets to the host. this will probably
2174          * generate destination unreachable messages from the host OS, but
2175          * the packets will be handled on SMBUS
2176          */
2177         manc |= E1000_MANC_EN_MNG2HOST;
2178         manc2h = er32(MANC2H);
2179 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2180 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2181         manc2h |= E1000_MNG2HOST_PORT_623;
2182         manc2h |= E1000_MNG2HOST_PORT_664;
2183         ew32(MANC2H, manc2h);
2184         ew32(MANC, manc);
2185 }
2186
2187 /**
2188  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2189  * @adapter: board private structure
2190  *
2191  * Configure the Tx unit of the MAC after a reset.
2192  **/
2193 static void e1000_configure_tx(struct e1000_adapter *adapter)
2194 {
2195         struct e1000_hw *hw = &adapter->hw;
2196         struct e1000_ring *tx_ring = adapter->tx_ring;
2197         u64 tdba;
2198         u32 tdlen, tctl, tipg, tarc;
2199         u32 ipgr1, ipgr2;
2200
2201         /* Setup the HW Tx Head and Tail descriptor pointers */
2202         tdba = tx_ring->dma;
2203         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2204         ew32(TDBAL, (tdba & DMA_32BIT_MASK));
2205         ew32(TDBAH, (tdba >> 32));
2206         ew32(TDLEN, tdlen);
2207         ew32(TDH, 0);
2208         ew32(TDT, 0);
2209         tx_ring->head = E1000_TDH;
2210         tx_ring->tail = E1000_TDT;
2211
2212         /* Set the default values for the Tx Inter Packet Gap timer */
2213         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2214         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2215         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2216
2217         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2218                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2219
2220         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2221         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2222         ew32(TIPG, tipg);
2223
2224         /* Set the Tx Interrupt Delay register */
2225         ew32(TIDV, adapter->tx_int_delay);
2226         /* Tx irq moderation */
2227         ew32(TADV, adapter->tx_abs_int_delay);
2228
2229         /* Program the Transmit Control Register */
2230         tctl = er32(TCTL);
2231         tctl &= ~E1000_TCTL_CT;
2232         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2233                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2234
2235         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2236                 tarc = er32(TARC(0));
2237                 /*
2238                  * set the speed mode bit, we'll clear it if we're not at
2239                  * gigabit link later
2240                  */
2241 #define SPEED_MODE_BIT (1 << 21)
2242                 tarc |= SPEED_MODE_BIT;
2243                 ew32(TARC(0), tarc);
2244         }
2245
2246         /* errata: program both queues to unweighted RR */
2247         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2248                 tarc = er32(TARC(0));
2249                 tarc |= 1;
2250                 ew32(TARC(0), tarc);
2251                 tarc = er32(TARC(1));
2252                 tarc |= 1;
2253                 ew32(TARC(1), tarc);
2254         }
2255
2256         e1000e_config_collision_dist(hw);
2257
2258         /* Setup Transmit Descriptor Settings for eop descriptor */
2259         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2260
2261         /* only set IDE if we are delaying interrupts using the timers */
2262         if (adapter->tx_int_delay)
2263                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2264
2265         /* enable Report Status bit */
2266         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2267
2268         ew32(TCTL, tctl);
2269
2270         adapter->tx_queue_len = adapter->netdev->tx_queue_len;
2271 }
2272
2273 /**
2274  * e1000_setup_rctl - configure the receive control registers
2275  * @adapter: Board private structure
2276  **/
2277 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2278                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2279 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2280 {
2281         struct e1000_hw *hw = &adapter->hw;
2282         u32 rctl, rfctl;
2283         u32 psrctl = 0;
2284         u32 pages = 0;
2285
2286         /* Program MC offset vector base */
2287         rctl = er32(RCTL);
2288         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2289         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2290                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2291                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2292
2293         /* Do not Store bad packets */
2294         rctl &= ~E1000_RCTL_SBP;
2295
2296         /* Enable Long Packet receive */
2297         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2298                 rctl &= ~E1000_RCTL_LPE;
2299         else
2300                 rctl |= E1000_RCTL_LPE;
2301
2302         /* Enable hardware CRC frame stripping */
2303         rctl |= E1000_RCTL_SECRC;
2304
2305         /* Setup buffer sizes */
2306         rctl &= ~E1000_RCTL_SZ_4096;
2307         rctl |= E1000_RCTL_BSEX;
2308         switch (adapter->rx_buffer_len) {
2309         case 256:
2310                 rctl |= E1000_RCTL_SZ_256;
2311                 rctl &= ~E1000_RCTL_BSEX;
2312                 break;
2313         case 512:
2314                 rctl |= E1000_RCTL_SZ_512;
2315                 rctl &= ~E1000_RCTL_BSEX;
2316                 break;
2317         case 1024:
2318                 rctl |= E1000_RCTL_SZ_1024;
2319                 rctl &= ~E1000_RCTL_BSEX;
2320                 break;
2321         case 2048:
2322         default:
2323                 rctl |= E1000_RCTL_SZ_2048;
2324                 rctl &= ~E1000_RCTL_BSEX;
2325                 break;
2326         case 4096:
2327                 rctl |= E1000_RCTL_SZ_4096;
2328                 break;
2329         case 8192:
2330                 rctl |= E1000_RCTL_SZ_8192;
2331                 break;
2332         case 16384:
2333                 rctl |= E1000_RCTL_SZ_16384;
2334                 break;
2335         }
2336
2337         /*
2338          * 82571 and greater support packet-split where the protocol
2339          * header is placed in skb->data and the packet data is
2340          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2341          * In the case of a non-split, skb->data is linearly filled,
2342          * followed by the page buffers.  Therefore, skb->data is
2343          * sized to hold the largest protocol header.
2344          *
2345          * allocations using alloc_page take too long for regular MTU
2346          * so only enable packet split for jumbo frames
2347          *
2348          * Using pages when the page size is greater than 16k wastes
2349          * a lot of memory, since we allocate 3 pages at all times
2350          * per packet.
2351          */
2352         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2353         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2354             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2355                 adapter->rx_ps_pages = pages;
2356         else
2357                 adapter->rx_ps_pages = 0;
2358
2359         if (adapter->rx_ps_pages) {
2360                 /* Configure extra packet-split registers */
2361                 rfctl = er32(RFCTL);
2362                 rfctl |= E1000_RFCTL_EXTEN;
2363                 /*
2364                  * disable packet split support for IPv6 extension headers,
2365                  * because some malformed IPv6 headers can hang the Rx
2366                  */
2367                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2368                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2369
2370                 ew32(RFCTL, rfctl);
2371
2372                 /* Enable Packet split descriptors */
2373                 rctl |= E1000_RCTL_DTYP_PS;
2374
2375                 psrctl |= adapter->rx_ps_bsize0 >>
2376                         E1000_PSRCTL_BSIZE0_SHIFT;
2377
2378                 switch (adapter->rx_ps_pages) {
2379                 case 3:
2380                         psrctl |= PAGE_SIZE <<
2381                                 E1000_PSRCTL_BSIZE3_SHIFT;
2382                 case 2:
2383                         psrctl |= PAGE_SIZE <<
2384                                 E1000_PSRCTL_BSIZE2_SHIFT;
2385                 case 1:
2386                         psrctl |= PAGE_SIZE >>
2387                                 E1000_PSRCTL_BSIZE1_SHIFT;
2388                         break;
2389                 }
2390
2391                 ew32(PSRCTL, psrctl);
2392         }
2393
2394         ew32(RCTL, rctl);
2395         /* just started the receive unit, no need to restart */
2396         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2397 }
2398
2399 /**
2400  * e1000_configure_rx - Configure Receive Unit after Reset
2401  * @adapter: board private structure
2402  *
2403  * Configure the Rx unit of the MAC after a reset.
2404  **/
2405 static void e1000_configure_rx(struct e1000_adapter *adapter)
2406 {
2407         struct e1000_hw *hw = &adapter->hw;
2408         struct e1000_ring *rx_ring = adapter->rx_ring;
2409         u64 rdba;
2410         u32 rdlen, rctl, rxcsum, ctrl_ext;
2411
2412         if (adapter->rx_ps_pages) {
2413                 /* this is a 32 byte descriptor */
2414                 rdlen = rx_ring->count *
2415                         sizeof(union e1000_rx_desc_packet_split);
2416                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2417                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2418         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2419                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2420                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2421                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2422         } else {
2423                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2424                 adapter->clean_rx = e1000_clean_rx_irq;
2425                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2426         }
2427
2428         /* disable receives while setting up the descriptors */
2429         rctl = er32(RCTL);
2430         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2431         e1e_flush();
2432         msleep(10);
2433
2434         /* set the Receive Delay Timer Register */
2435         ew32(RDTR, adapter->rx_int_delay);
2436
2437         /* irq moderation */
2438         ew32(RADV, adapter->rx_abs_int_delay);
2439         if (adapter->itr_setting != 0)
2440                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2441
2442         ctrl_ext = er32(CTRL_EXT);
2443         /* Reset delay timers after every interrupt */
2444         ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
2445         /* Auto-Mask interrupts upon ICR access */
2446         ctrl_ext |= E1000_CTRL_EXT_IAME;
2447         ew32(IAM, 0xffffffff);
2448         ew32(CTRL_EXT, ctrl_ext);
2449         e1e_flush();
2450
2451         /*
2452          * Setup the HW Rx Head and Tail Descriptor Pointers and
2453          * the Base and Length of the Rx Descriptor Ring
2454          */
2455         rdba = rx_ring->dma;
2456         ew32(RDBAL, (rdba & DMA_32BIT_MASK));
2457         ew32(RDBAH, (rdba >> 32));
2458         ew32(RDLEN, rdlen);
2459         ew32(RDH, 0);
2460         ew32(RDT, 0);
2461         rx_ring->head = E1000_RDH;
2462         rx_ring->tail = E1000_RDT;
2463
2464         /* Enable Receive Checksum Offload for TCP and UDP */
2465         rxcsum = er32(RXCSUM);
2466         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2467                 rxcsum |= E1000_RXCSUM_TUOFL;
2468
2469                 /*
2470                  * IPv4 payload checksum for UDP fragments must be
2471                  * used in conjunction with packet-split.
2472                  */
2473                 if (adapter->rx_ps_pages)
2474                         rxcsum |= E1000_RXCSUM_IPPCSE;
2475         } else {
2476                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2477                 /* no need to clear IPPCSE as it defaults to 0 */
2478         }
2479         ew32(RXCSUM, rxcsum);
2480
2481         /*
2482          * Enable early receives on supported devices, only takes effect when
2483          * packet size is equal or larger than the specified value (in 8 byte
2484          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2485          */
2486         if ((adapter->flags & FLAG_HAS_ERT) &&
2487             (adapter->netdev->mtu > ETH_DATA_LEN)) {
2488                 u32 rxdctl = er32(RXDCTL(0));
2489                 ew32(RXDCTL(0), rxdctl | 0x3);
2490                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2491                 /*
2492                  * With jumbo frames and early-receive enabled, excessive
2493                  * C4->C2 latencies result in dropped transactions.
2494                  */
2495                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2496                                           e1000e_driver_name, 55);
2497         } else {
2498                 pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2499                                           e1000e_driver_name,
2500                                           PM_QOS_DEFAULT_VALUE);
2501         }
2502
2503         /* Enable Receives */
2504         ew32(RCTL, rctl);
2505 }
2506
2507 /**
2508  *  e1000_update_mc_addr_list - Update Multicast addresses
2509  *  @hw: pointer to the HW structure
2510  *  @mc_addr_list: array of multicast addresses to program
2511  *  @mc_addr_count: number of multicast addresses to program
2512  *  @rar_used_count: the first RAR register free to program
2513  *  @rar_count: total number of supported Receive Address Registers
2514  *
2515  *  Updates the Receive Address Registers and Multicast Table Array.
2516  *  The caller must have a packed mc_addr_list of multicast addresses.
2517  *  The parameter rar_count will usually be hw->mac.rar_entry_count
2518  *  unless there are workarounds that change this.  Currently no func pointer
2519  *  exists and all implementations are handled in the generic version of this
2520  *  function.
2521  **/
2522 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2523                                       u32 mc_addr_count, u32 rar_used_count,
2524                                       u32 rar_count)
2525 {
2526         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2527                                         rar_used_count, rar_count);
2528 }
2529
2530 /**
2531  * e1000_set_multi - Multicast and Promiscuous mode set
2532  * @netdev: network interface device structure
2533  *
2534  * The set_multi entry point is called whenever the multicast address
2535  * list or the network interface flags are updated.  This routine is
2536  * responsible for configuring the hardware for proper multicast,
2537  * promiscuous mode, and all-multi behavior.
2538  **/
2539 static void e1000_set_multi(struct net_device *netdev)
2540 {
2541         struct e1000_adapter *adapter = netdev_priv(netdev);
2542         struct e1000_hw *hw = &adapter->hw;
2543         struct e1000_mac_info *mac = &hw->mac;
2544         struct dev_mc_list *mc_ptr;
2545         u8  *mta_list;
2546         u32 rctl;
2547         int i;
2548
2549         /* Check for Promiscuous and All Multicast modes */
2550
2551         rctl = er32(RCTL);
2552
2553         if (netdev->flags & IFF_PROMISC) {
2554                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2555                 rctl &= ~E1000_RCTL_VFE;
2556         } else {
2557                 if (netdev->flags & IFF_ALLMULTI) {
2558                         rctl |= E1000_RCTL_MPE;
2559                         rctl &= ~E1000_RCTL_UPE;
2560                 } else {
2561                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2562                 }
2563                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2564                         rctl |= E1000_RCTL_VFE;
2565         }
2566
2567         ew32(RCTL, rctl);
2568
2569         if (netdev->mc_count) {
2570                 mta_list = kmalloc(netdev->mc_count * 6, GFP_ATOMIC);
2571                 if (!mta_list)
2572                         return;
2573
2574                 /* prepare a packed array of only addresses. */
2575                 mc_ptr = netdev->mc_list;
2576
2577                 for (i = 0; i < netdev->mc_count; i++) {
2578                         if (!mc_ptr)
2579                                 break;
2580                         memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr,
2581                                ETH_ALEN);
2582                         mc_ptr = mc_ptr->next;
2583                 }
2584
2585                 e1000_update_mc_addr_list(hw, mta_list, i, 1,
2586                                           mac->rar_entry_count);
2587                 kfree(mta_list);
2588         } else {
2589                 /*
2590                  * if we're called from probe, we might not have
2591                  * anything to do here, so clear out the list
2592                  */
2593                 e1000_update_mc_addr_list(hw, NULL, 0, 1, mac->rar_entry_count);
2594         }
2595 }
2596
2597 /**
2598  * e1000_configure - configure the hardware for Rx and Tx
2599  * @adapter: private board structure
2600  **/
2601 static void e1000_configure(struct e1000_adapter *adapter)
2602 {
2603         e1000_set_multi(adapter->netdev);
2604
2605         e1000_restore_vlan(adapter);
2606         e1000_init_manageability(adapter);
2607
2608         e1000_configure_tx(adapter);
2609         e1000_setup_rctl(adapter);
2610         e1000_configure_rx(adapter);
2611         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2612 }
2613
2614 /**
2615  * e1000e_power_up_phy - restore link in case the phy was powered down
2616  * @adapter: address of board private structure
2617  *
2618  * The phy may be powered down to save power and turn off link when the
2619  * driver is unloaded and wake on lan is not enabled (among others)
2620  * *** this routine MUST be followed by a call to e1000e_reset ***
2621  **/
2622 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2623 {
2624         u16 mii_reg = 0;
2625
2626         /* Just clear the power down bit to wake the phy back up */
2627         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
2628                 /*
2629                  * According to the manual, the phy will retain its
2630                  * settings across a power-down/up cycle
2631                  */
2632                 e1e_rphy(&adapter->hw, PHY_CONTROL, &mii_reg);
2633                 mii_reg &= ~MII_CR_POWER_DOWN;
2634                 e1e_wphy(&adapter->hw, PHY_CONTROL, mii_reg);
2635         }
2636
2637         adapter->hw.mac.ops.setup_link(&adapter->hw);
2638 }
2639
2640 /**
2641  * e1000_power_down_phy - Power down the PHY
2642  *
2643  * Power down the PHY so no link is implied when interface is down
2644  * The PHY cannot be powered down is management or WoL is active
2645  */
2646 static void e1000_power_down_phy(struct e1000_adapter *adapter)
2647 {
2648         struct e1000_hw *hw = &adapter->hw;
2649         u16 mii_reg;
2650
2651         /* WoL is enabled */
2652         if (adapter->wol)
2653                 return;
2654
2655         /* non-copper PHY? */
2656         if (adapter->hw.phy.media_type != e1000_media_type_copper)
2657                 return;
2658
2659         /* reset is blocked because of a SoL/IDER session */
2660         if (e1000e_check_mng_mode(hw) || e1000_check_reset_block(hw))
2661                 return;
2662
2663         /* manageability (AMT) is enabled */
2664         if (er32(MANC) & E1000_MANC_SMBUS_EN)
2665                 return;
2666
2667         /* power down the PHY */
2668         e1e_rphy(hw, PHY_CONTROL, &mii_reg);
2669         mii_reg |= MII_CR_POWER_DOWN;
2670         e1e_wphy(hw, PHY_CONTROL, mii_reg);
2671         mdelay(1);
2672 }
2673
2674 /**
2675  * e1000e_reset - bring the hardware into a known good state
2676  *
2677  * This function boots the hardware and enables some settings that
2678  * require a configuration cycle of the hardware - those cannot be
2679  * set/changed during runtime. After reset the device needs to be
2680  * properly configured for Rx, Tx etc.
2681  */
2682 void e1000e_reset(struct e1000_adapter *adapter)
2683 {
2684         struct e1000_mac_info *mac = &adapter->hw.mac;
2685         struct e1000_fc_info *fc = &adapter->hw.fc;
2686         struct e1000_hw *hw = &adapter->hw;
2687         u32 tx_space, min_tx_space, min_rx_space;
2688         u32 pba = adapter->pba;
2689         u16 hwm;
2690
2691         /* reset Packet Buffer Allocation to default */
2692         ew32(PBA, pba);
2693
2694         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
2695                 /*
2696                  * To maintain wire speed transmits, the Tx FIFO should be
2697                  * large enough to accommodate two full transmit packets,
2698                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2699                  * the Rx FIFO should be large enough to accommodate at least
2700                  * one full receive packet and is similarly rounded up and
2701                  * expressed in KB.
2702                  */
2703                 pba = er32(PBA);
2704                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2705                 tx_space = pba >> 16;
2706                 /* lower 16 bits has Rx packet buffer allocation size in KB */
2707                 pba &= 0xffff;
2708                 /*
2709                  * the Tx fifo also stores 16 bytes of information about the tx
2710                  * but don't include ethernet FCS because hardware appends it
2711                  */
2712                 min_tx_space = (adapter->max_frame_size +
2713                                 sizeof(struct e1000_tx_desc) -
2714                                 ETH_FCS_LEN) * 2;
2715                 min_tx_space = ALIGN(min_tx_space, 1024);
2716                 min_tx_space >>= 10;
2717                 /* software strips receive CRC, so leave room for it */
2718                 min_rx_space = adapter->max_frame_size;
2719                 min_rx_space = ALIGN(min_rx_space, 1024);
2720                 min_rx_space >>= 10;
2721
2722                 /*
2723                  * If current Tx allocation is less than the min Tx FIFO size,
2724                  * and the min Tx FIFO size is less than the current Rx FIFO
2725                  * allocation, take space away from current Rx allocation
2726                  */
2727                 if ((tx_space < min_tx_space) &&
2728                     ((min_tx_space - tx_space) < pba)) {
2729                         pba -= min_tx_space - tx_space;
2730
2731                         /*
2732                          * if short on Rx space, Rx wins and must trump tx
2733                          * adjustment or use Early Receive if available
2734                          */
2735                         if ((pba < min_rx_space) &&
2736                             (!(adapter->flags & FLAG_HAS_ERT)))
2737                                 /* ERT enabled in e1000_configure_rx */
2738                                 pba = min_rx_space;
2739                 }
2740
2741                 ew32(PBA, pba);
2742         }
2743
2744
2745         /*
2746          * flow control settings
2747          *
2748          * The high water mark must be low enough to fit one full frame
2749          * (or the size used for early receive) above it in the Rx FIFO.
2750          * Set it to the lower of:
2751          * - 90% of the Rx FIFO size, and
2752          * - the full Rx FIFO size minus the early receive size (for parts
2753          *   with ERT support assuming ERT set to E1000_ERT_2048), or
2754          * - the full Rx FIFO size minus one full frame
2755          */
2756         if (adapter->flags & FLAG_HAS_ERT)
2757                 hwm = min(((pba << 10) * 9 / 10),
2758                           ((pba << 10) - (E1000_ERT_2048 << 3)));
2759         else
2760                 hwm = min(((pba << 10) * 9 / 10),
2761                           ((pba << 10) - adapter->max_frame_size));
2762
2763         fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
2764         fc->low_water = fc->high_water - 8;
2765
2766         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
2767                 fc->pause_time = 0xFFFF;
2768         else
2769                 fc->pause_time = E1000_FC_PAUSE_TIME;
2770         fc->send_xon = 1;
2771         fc->type = fc->original_type;
2772
2773         /* Allow time for pending master requests to run */
2774         mac->ops.reset_hw(hw);
2775
2776         /*
2777          * For parts with AMT enabled, let the firmware know
2778          * that the network interface is in control
2779          */
2780         if (adapter->flags & FLAG_HAS_AMT)
2781                 e1000_get_hw_control(adapter);
2782
2783         ew32(WUC, 0);
2784
2785         if (mac->ops.init_hw(hw))
2786                 e_err("Hardware Error\n");
2787
2788         e1000_update_mng_vlan(adapter);
2789
2790         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2791         ew32(VET, ETH_P_8021Q);
2792
2793         e1000e_reset_adaptive(hw);
2794         e1000_get_phy_info(hw);
2795
2796         if (!(adapter->flags & FLAG_SMART_POWER_DOWN)) {
2797                 u16 phy_data = 0;
2798                 /*
2799                  * speed up time to link by disabling smart power down, ignore
2800                  * the return value of this function because there is nothing
2801                  * different we would do if it failed
2802                  */
2803                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
2804                 phy_data &= ~IGP02E1000_PM_SPD;
2805                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
2806         }
2807 }
2808
2809 int e1000e_up(struct e1000_adapter *adapter)
2810 {
2811         struct e1000_hw *hw = &adapter->hw;
2812
2813         /* hardware has been reset, we need to reload some things */
2814         e1000_configure(adapter);
2815
2816         clear_bit(__E1000_DOWN, &adapter->state);
2817
2818         napi_enable(&adapter->napi);
2819         if (adapter->msix_entries)
2820                 e1000_configure_msix(adapter);
2821         e1000_irq_enable(adapter);
2822
2823         /* fire a link change interrupt to start the watchdog */
2824         ew32(ICS, E1000_ICS_LSC);
2825         return 0;
2826 }
2827
2828 void e1000e_down(struct e1000_adapter *adapter)
2829 {
2830         struct net_device *netdev = adapter->netdev;
2831         struct e1000_hw *hw = &adapter->hw;
2832         u32 tctl, rctl;
2833
2834         /*
2835          * signal that we're down so the interrupt handler does not
2836          * reschedule our watchdog timer
2837          */
2838         set_bit(__E1000_DOWN, &adapter->state);
2839
2840         /* disable receives in the hardware */
2841         rctl = er32(RCTL);
2842         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2843         /* flush and sleep below */
2844
2845         netif_tx_stop_all_queues(netdev);
2846
2847         /* disable transmits in the hardware */
2848         tctl = er32(TCTL);
2849         tctl &= ~E1000_TCTL_EN;
2850         ew32(TCTL, tctl);
2851         /* flush both disables and wait for them to finish */
2852         e1e_flush();
2853         msleep(10);
2854
2855         napi_disable(&adapter->napi);
2856         e1000_irq_disable(adapter);
2857
2858         del_timer_sync(&adapter->watchdog_timer);
2859         del_timer_sync(&adapter->phy_info_timer);
2860
2861         netdev->tx_queue_len = adapter->tx_queue_len;
2862         netif_carrier_off(netdev);
2863         adapter->link_speed = 0;
2864         adapter->link_duplex = 0;
2865
2866         if (!pci_channel_offline(adapter->pdev))
2867                 e1000e_reset(adapter);
2868         e1000_clean_tx_ring(adapter);
2869         e1000_clean_rx_ring(adapter);
2870
2871         /*
2872          * TODO: for power management, we could drop the link and
2873          * pci_disable_device here.
2874          */
2875 }
2876
2877 void e1000e_reinit_locked(struct e1000_adapter *adapter)
2878 {
2879         might_sleep();
2880         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
2881                 msleep(1);
2882         e1000e_down(adapter);
2883         e1000e_up(adapter);
2884         clear_bit(__E1000_RESETTING, &adapter->state);
2885 }
2886
2887 /**
2888  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
2889  * @adapter: board private structure to initialize
2890  *
2891  * e1000_sw_init initializes the Adapter private data structure.
2892  * Fields are initialized based on PCI device information and
2893  * OS network device settings (MTU size).
2894  **/
2895 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
2896 {
2897         struct net_device *netdev = adapter->netdev;
2898
2899         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
2900         adapter->rx_ps_bsize0 = 128;
2901         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2902         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2903
2904         e1000e_set_interrupt_capability(adapter);
2905
2906         if (e1000_alloc_queues(adapter))
2907                 return -ENOMEM;
2908
2909         spin_lock_init(&adapter->tx_queue_lock);
2910
2911         /* Explicitly disable IRQ since the NIC can be in any state. */
2912         e1000_irq_disable(adapter);
2913
2914         set_bit(__E1000_DOWN, &adapter->state);
2915         return 0;
2916 }
2917
2918 /**
2919  * e1000_intr_msi_test - Interrupt Handler
2920  * @irq: interrupt number
2921  * @data: pointer to a network interface device structure
2922  **/
2923 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
2924 {
2925         struct net_device *netdev = data;
2926         struct e1000_adapter *adapter = netdev_priv(netdev);
2927         struct e1000_hw *hw = &adapter->hw;
2928         u32 icr = er32(ICR);
2929
2930         e_dbg("%s: icr is %08X\n", netdev->name, icr);
2931         if (icr & E1000_ICR_RXSEQ) {
2932                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
2933                 wmb();
2934         }
2935
2936         return IRQ_HANDLED;
2937 }
2938
2939 /**
2940  * e1000_test_msi_interrupt - Returns 0 for successful test
2941  * @adapter: board private struct
2942  *
2943  * code flow taken from tg3.c
2944  **/
2945 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
2946 {
2947         struct net_device *netdev = adapter->netdev;
2948         struct e1000_hw *hw = &adapter->hw;
2949         int err;
2950
2951         /* poll_enable hasn't been called yet, so don't need disable */
2952         /* clear any pending events */
2953         er32(ICR);
2954
2955         /* free the real vector and request a test handler */
2956         e1000_free_irq(adapter);
2957         e1000e_reset_interrupt_capability(adapter);
2958
2959         /* Assume that the test fails, if it succeeds then the test
2960          * MSI irq handler will unset this flag */
2961         adapter->flags |= FLAG_MSI_TEST_FAILED;
2962
2963         err = pci_enable_msi(adapter->pdev);
2964         if (err)
2965                 goto msi_test_failed;
2966
2967         err = request_irq(adapter->pdev->irq, &e1000_intr_msi_test, 0,
2968                           netdev->name, netdev);
2969         if (err) {
2970                 pci_disable_msi(adapter->pdev);
2971                 goto msi_test_failed;
2972         }
2973
2974         wmb();
2975
2976         e1000_irq_enable(adapter);
2977
2978         /* fire an unusual interrupt on the test handler */
2979         ew32(ICS, E1000_ICS_RXSEQ);
2980         e1e_flush();
2981         msleep(50);
2982
2983         e1000_irq_disable(adapter);
2984
2985         rmb();
2986
2987         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
2988                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2989                 err = -EIO;
2990                 e_info("MSI interrupt test failed!\n");
2991         }
2992
2993         free_irq(adapter->pdev->irq, netdev);
2994         pci_disable_msi(adapter->pdev);
2995
2996         if (err == -EIO)
2997                 goto msi_test_failed;
2998
2999         /* okay so the test worked, restore settings */
3000         e_dbg("%s: MSI interrupt test succeeded!\n", netdev->name);
3001 msi_test_failed:
3002         e1000e_set_interrupt_capability(adapter);
3003         e1000_request_irq(adapter);
3004         return err;
3005 }
3006
3007 /**
3008  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3009  * @adapter: board private struct
3010  *
3011  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3012  **/
3013 static int e1000_test_msi(struct e1000_adapter *adapter)
3014 {
3015         int err;
3016         u16 pci_cmd;
3017
3018         if (!(adapter->flags & FLAG_MSI_ENABLED))
3019                 return 0;
3020
3021         /* disable SERR in case the MSI write causes a master abort */
3022         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3023         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3024                               pci_cmd & ~PCI_COMMAND_SERR);
3025
3026         err = e1000_test_msi_interrupt(adapter);
3027
3028         /* restore previous setting of command word */
3029         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3030
3031         /* success ! */
3032         if (!err)
3033                 return 0;
3034
3035         /* EIO means MSI test failed */
3036         if (err != -EIO)
3037                 return err;
3038
3039         /* back to INTx mode */
3040         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3041
3042         e1000_free_irq(adapter);
3043
3044         err = e1000_request_irq(adapter);
3045
3046         return err;
3047 }
3048
3049 /**
3050  * e1000_open - Called when a network interface is made active
3051  * @netdev: network interface device structure
3052  *
3053  * Returns 0 on success, negative value on failure
3054  *
3055  * The open entry point is called when a network interface is made
3056  * active by the system (IFF_UP).  At this point all resources needed
3057  * for transmit and receive operations are allocated, the interrupt
3058  * handler is registered with the OS, the watchdog timer is started,
3059  * and the stack is notified that the interface is ready.
3060  **/
3061 static int e1000_open(struct net_device *netdev)
3062 {
3063         struct e1000_adapter *adapter = netdev_priv(netdev);
3064         struct e1000_hw *hw = &adapter->hw;
3065         int err;
3066
3067         /* disallow open during test */
3068         if (test_bit(__E1000_TESTING, &adapter->state))
3069                 return -EBUSY;
3070
3071         /* allocate transmit descriptors */
3072         err = e1000e_setup_tx_resources(adapter);
3073         if (err)
3074                 goto err_setup_tx;
3075
3076         /* allocate receive descriptors */
3077         err = e1000e_setup_rx_resources(adapter);
3078         if (err)
3079                 goto err_setup_rx;
3080
3081         e1000e_power_up_phy(adapter);
3082
3083         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3084         if ((adapter->hw.mng_cookie.status &
3085              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3086                 e1000_update_mng_vlan(adapter);
3087
3088         /*
3089          * If AMT is enabled, let the firmware know that the network
3090          * interface is now open
3091          */
3092         if (adapter->flags & FLAG_HAS_AMT)
3093                 e1000_get_hw_control(adapter);
3094
3095         /*
3096          * before we allocate an interrupt, we must be ready to handle it.
3097          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3098          * as soon as we call pci_request_irq, so we have to setup our
3099          * clean_rx handler before we do so.
3100          */
3101         e1000_configure(adapter);
3102
3103         err = e1000_request_irq(adapter);
3104         if (err)
3105                 goto err_req_irq;
3106
3107         /*
3108          * Work around PCIe errata with MSI interrupts causing some chipsets to
3109          * ignore e1000e MSI messages, which means we need to test our MSI
3110          * interrupt now
3111          */
3112         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3113                 err = e1000_test_msi(adapter);
3114                 if (err) {
3115                         e_err("Interrupt allocation failed\n");
3116                         goto err_req_irq;
3117                 }
3118         }
3119
3120         /* From here on the code is the same as e1000e_up() */
3121         clear_bit(__E1000_DOWN, &adapter->state);
3122
3123         napi_enable(&adapter->napi);
3124
3125         e1000_irq_enable(adapter);
3126
3127         netif_tx_start_all_queues(netdev);
3128
3129         /* fire a link status change interrupt to start the watchdog */
3130         ew32(ICS, E1000_ICS_LSC);
3131
3132         return 0;
3133
3134 err_req_irq:
3135         e1000_release_hw_control(adapter);
3136         e1000_power_down_phy(adapter);
3137         e1000e_free_rx_resources(adapter);
3138 err_setup_rx:
3139         e1000e_free_tx_resources(adapter);
3140 err_setup_tx:
3141         e1000e_reset(adapter);
3142
3143         return err;
3144 }
3145
3146 /**
3147  * e1000_close - Disables a network interface
3148  * @netdev: network interface device structure
3149  *
3150  * Returns 0, this is not allowed to fail
3151  *
3152  * The close entry point is called when an interface is de-activated
3153  * by the OS.  The hardware is still under the drivers control, but
3154  * needs to be disabled.  A global MAC reset is issued to stop the
3155  * hardware, and all transmit and receive resources are freed.
3156  **/
3157 static int e1000_close(struct net_device *netdev)
3158 {
3159         struct e1000_adapter *adapter = netdev_priv(netdev);
3160
3161         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3162         e1000e_down(adapter);
3163         e1000_power_down_phy(adapter);
3164         e1000_free_irq(adapter);
3165
3166         e1000e_free_tx_resources(adapter);
3167         e1000e_free_rx_resources(adapter);
3168
3169         /*
3170          * kill manageability vlan ID if supported, but not if a vlan with
3171          * the same ID is registered on the host OS (let 8021q kill it)
3172          */
3173         if ((adapter->hw.mng_cookie.status &
3174                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3175              !(adapter->vlgrp &&
3176                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3177                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3178
3179         /*
3180          * If AMT is enabled, let the firmware know that the network
3181          * interface is now closed
3182          */
3183         if (adapter->flags & FLAG_HAS_AMT)
3184                 e1000_release_hw_control(adapter);
3185
3186         return 0;
3187 }
3188 /**
3189  * e1000_set_mac - Change the Ethernet Address of the NIC
3190  * @netdev: network interface device structure
3191  * @p: pointer to an address structure
3192  *
3193  * Returns 0 on success, negative on failure
3194  **/
3195 static int e1000_set_mac(struct net_device *netdev, void *p)
3196 {
3197         struct e1000_adapter *adapter = netdev_priv(netdev);
3198         struct sockaddr *addr = p;
3199
3200         if (!is_valid_ether_addr(addr->sa_data))
3201                 return -EADDRNOTAVAIL;
3202
3203         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3204         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3205
3206         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3207
3208         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3209                 /* activate the work around */
3210                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3211
3212                 /*
3213                  * Hold a copy of the LAA in RAR[14] This is done so that
3214                  * between the time RAR[0] gets clobbered  and the time it
3215                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3216                  * of the RARs and no incoming packets directed to this port
3217                  * are dropped. Eventually the LAA will be in RAR[0] and
3218                  * RAR[14]
3219                  */
3220                 e1000e_rar_set(&adapter->hw,
3221                               adapter->hw.mac.addr,
3222                               adapter->hw.mac.rar_entry_count - 1);
3223         }
3224
3225         return 0;
3226 }
3227
3228 /**
3229  * e1000e_update_phy_task - work thread to update phy
3230  * @work: pointer to our work struct
3231  *
3232  * this worker thread exists because we must acquire a
3233  * semaphore to read the phy, which we could msleep while
3234  * waiting for it, and we can't msleep in a timer.
3235  **/
3236 static void e1000e_update_phy_task(struct work_struct *work)
3237 {
3238         struct e1000_adapter *adapter = container_of(work,
3239                                         struct e1000_adapter, update_phy_task);
3240         e1000_get_phy_info(&adapter->hw);
3241 }
3242
3243 /*
3244  * Need to wait a few seconds after link up to get diagnostic information from
3245  * the phy
3246  */
3247 static void e1000_update_phy_info(unsigned long data)
3248 {
3249         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3250         schedule_work(&adapter->update_phy_task);
3251 }
3252
3253 /**
3254  * e1000e_update_stats - Update the board statistics counters
3255  * @adapter: board private structure
3256  **/
3257 void e1000e_update_stats(struct e1000_adapter *adapter)
3258 {
3259         struct e1000_hw *hw = &adapter->hw;
3260         struct pci_dev *pdev = adapter->pdev;
3261
3262         /*
3263          * Prevent stats update while adapter is being reset, or if the pci
3264          * connection is down.
3265          */
3266         if (adapter->link_speed == 0)
3267                 return;
3268         if (pci_channel_offline(pdev))
3269                 return;
3270
3271         adapter->stats.crcerrs += er32(CRCERRS);
3272         adapter->stats.gprc += er32(GPRC);
3273         adapter->stats.gorc += er32(GORCL);
3274         er32(GORCH); /* Clear gorc */
3275         adapter->stats.bprc += er32(BPRC);
3276         adapter->stats.mprc += er32(MPRC);
3277         adapter->stats.roc += er32(ROC);
3278
3279         adapter->stats.mpc += er32(MPC);
3280         adapter->stats.scc += er32(SCC);
3281         adapter->stats.ecol += er32(ECOL);
3282         adapter->stats.mcc += er32(MCC);
3283         adapter->stats.latecol += er32(LATECOL);
3284         adapter->stats.dc += er32(DC);
3285         adapter->stats.xonrxc += er32(XONRXC);
3286         adapter->stats.xontxc += er32(XONTXC);
3287         adapter->stats.xoffrxc += er32(XOFFRXC);
3288         adapter->stats.xofftxc += er32(XOFFTXC);
3289         adapter->stats.gptc += er32(GPTC);
3290         adapter->stats.gotc += er32(GOTCL);
3291         er32(GOTCH); /* Clear gotc */
3292         adapter->stats.rnbc += er32(RNBC);
3293         adapter->stats.ruc += er32(RUC);
3294
3295         adapter->stats.mptc += er32(MPTC);
3296         adapter->stats.bptc += er32(BPTC);
3297
3298         /* used for adaptive IFS */
3299
3300         hw->mac.tx_packet_delta = er32(TPT);
3301         adapter->stats.tpt += hw->mac.tx_packet_delta;
3302         hw->mac.collision_delta = er32(COLC);
3303         adapter->stats.colc += hw->mac.collision_delta;
3304
3305         adapter->stats.algnerrc += er32(ALGNERRC);
3306         adapter->stats.rxerrc += er32(RXERRC);
3307         if (hw->mac.type != e1000_82574)
3308                 adapter->stats.tncrs += er32(TNCRS);
3309         adapter->stats.cexterr += er32(CEXTERR);
3310         adapter->stats.tsctc += er32(TSCTC);
3311         adapter->stats.tsctfc += er32(TSCTFC);
3312
3313         /* Fill out the OS statistics structure */
3314         adapter->net_stats.multicast = adapter->stats.mprc;
3315         adapter->net_stats.collisions = adapter->stats.colc;
3316
3317         /* Rx Errors */
3318
3319         /*
3320          * RLEC on some newer hardware can be incorrect so build
3321          * our own version based on RUC and ROC
3322          */
3323         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3324                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3325                 adapter->stats.ruc + adapter->stats.roc +
3326                 adapter->stats.cexterr;
3327         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3328                                               adapter->stats.roc;
3329         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3330         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3331         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3332
3333         /* Tx Errors */
3334         adapter->net_stats.tx_errors = adapter->stats.ecol +
3335                                        adapter->stats.latecol;
3336         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3337         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3338         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3339
3340         /* Tx Dropped needs to be maintained elsewhere */
3341
3342         /* Management Stats */
3343         adapter->stats.mgptc += er32(MGTPTC);
3344         adapter->stats.mgprc += er32(MGTPRC);
3345         adapter->stats.mgpdc += er32(MGTPDC);
3346 }
3347
3348 /**
3349  * e1000_phy_read_status - Update the PHY register status snapshot
3350  * @adapter: board private structure
3351  **/
3352 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3353 {
3354         struct e1000_hw *hw = &adapter->hw;
3355         struct e1000_phy_regs *phy = &adapter->phy_regs;
3356         int ret_val;
3357
3358         if ((er32(STATUS) & E1000_STATUS_LU) &&
3359             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3360                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3361                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3362                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3363                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3364                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3365                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3366                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3367                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3368                 if (ret_val)
3369                         e_warn("Error reading PHY register\n");
3370         } else {
3371                 /*
3372                  * Do not read PHY registers if link is not up
3373                  * Set values to typical power-on defaults
3374                  */
3375                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3376                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3377                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3378                              BMSR_ERCAP);
3379                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3380                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3381                 phy->lpa = 0;
3382                 phy->expansion = EXPANSION_ENABLENPAGE;
3383                 phy->ctrl1000 = ADVERTISE_1000FULL;
3384                 phy->stat1000 = 0;
3385                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3386         }
3387 }
3388
3389 static void e1000_print_link_info(struct e1000_adapter *adapter)
3390 {
3391         struct e1000_hw *hw = &adapter->hw;
3392         u32 ctrl = er32(CTRL);
3393
3394         e_info("Link is Up %d Mbps %s, Flow Control: %s\n",
3395                adapter->link_speed,
3396                (adapter->link_duplex == FULL_DUPLEX) ?
3397                                 "Full Duplex" : "Half Duplex",
3398                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3399                                 "RX/TX" :
3400                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3401                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3402 }
3403
3404 static bool e1000_has_link(struct e1000_adapter *adapter)
3405 {
3406         struct e1000_hw *hw = &adapter->hw;
3407         bool link_active = 0;
3408         s32 ret_val = 0;
3409
3410         /*
3411          * get_link_status is set on LSC (link status) interrupt or
3412          * Rx sequence error interrupt.  get_link_status will stay
3413          * false until the check_for_link establishes link
3414          * for copper adapters ONLY
3415          */
3416         switch (hw->phy.media_type) {
3417         case e1000_media_type_copper:
3418                 if (hw->mac.get_link_status) {
3419                         ret_val = hw->mac.ops.check_for_link(hw);
3420                         link_active = !hw->mac.get_link_status;
3421                 } else {
3422                         link_active = 1;
3423                 }
3424                 break;
3425         case e1000_media_type_fiber:
3426                 ret_val = hw->mac.ops.check_for_link(hw);
3427                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3428                 break;
3429         case e1000_media_type_internal_serdes:
3430                 ret_val = hw->mac.ops.check_for_link(hw);
3431                 link_active = adapter->hw.mac.serdes_has_link;
3432                 break;
3433         default:
3434         case e1000_media_type_unknown:
3435                 break;
3436         }
3437
3438         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3439             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3440                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3441                 e_info("Gigabit has been disabled, downgrading speed\n");
3442         }
3443
3444         return link_active;
3445 }
3446
3447 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3448 {
3449         /* make sure the receive unit is started */
3450         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3451             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3452                 struct e1000_hw *hw = &adapter->hw;
3453                 u32 rctl = er32(RCTL);
3454                 ew32(RCTL, rctl | E1000_RCTL_EN);
3455                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3456         }
3457 }
3458
3459 /**
3460  * e1000_watchdog - Timer Call-back
3461  * @data: pointer to adapter cast into an unsigned long
3462  **/
3463 static void e1000_watchdog(unsigned long data)
3464 {
3465         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3466
3467         /* Do the rest outside of interrupt context */
3468         schedule_work(&adapter->watchdog_task);
3469
3470         /* TODO: make this use queue_delayed_work() */
3471 }
3472
3473 static void e1000_watchdog_task(struct work_struct *work)
3474 {
3475         struct e1000_adapter *adapter = container_of(work,
3476                                         struct e1000_adapter, watchdog_task);
3477         struct net_device *netdev = adapter->netdev;
3478         struct e1000_mac_info *mac = &adapter->hw.mac;
3479         struct e1000_ring *tx_ring = adapter->tx_ring;
3480         struct e1000_hw *hw = &adapter->hw;
3481         u32 link, tctl;
3482         int tx_pending = 0;
3483
3484         link = e1000_has_link(adapter);
3485         if ((netif_carrier_ok(netdev)) && link) {
3486                 e1000e_enable_receives(adapter);
3487                 goto link_up;
3488         }
3489
3490         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3491             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3492                 e1000_update_mng_vlan(adapter);
3493
3494         if (link) {
3495                 if (!netif_carrier_ok(netdev)) {
3496                         bool txb2b = 1;
3497                         /* update snapshot of PHY registers on LSC */
3498                         e1000_phy_read_status(adapter);
3499                         mac->ops.get_link_up_info(&adapter->hw,
3500                                                    &adapter->link_speed,
3501                                                    &adapter->link_duplex);
3502                         e1000_print_link_info(adapter);
3503                         /*
3504                          * On supported PHYs, check for duplex mismatch only
3505                          * if link has autonegotiated at 10/100 half
3506                          */
3507                         if ((hw->phy.type == e1000_phy_igp_3 ||
3508                              hw->phy.type == e1000_phy_bm) &&
3509                             (hw->mac.autoneg == true) &&
3510                             (adapter->link_speed == SPEED_10 ||
3511                              adapter->link_speed == SPEED_100) &&
3512                             (adapter->link_duplex == HALF_DUPLEX)) {
3513                                 u16 autoneg_exp;
3514
3515                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3516
3517                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3518                                         e_info("Autonegotiated half duplex but"
3519                                                " link partner cannot autoneg. "
3520                                                " Try forcing full duplex if "
3521                                                "link gets many collisions.\n");
3522                         }
3523
3524                         /*
3525                          * tweak tx_queue_len according to speed/duplex
3526                          * and adjust the timeout factor
3527                          */
3528                         netdev->tx_queue_len = adapter->tx_queue_len;
3529                         adapter->tx_timeout_factor = 1;
3530                         switch (adapter->link_speed) {
3531                         case SPEED_10:
3532                                 txb2b = 0;
3533                                 netdev->tx_queue_len = 10;
3534                                 adapter->tx_timeout_factor = 16;
3535                                 break;
3536                         case SPEED_100:
3537                                 txb2b = 0;
3538                                 netdev->tx_queue_len = 100;
3539                                 /* maybe add some timeout factor ? */
3540                                 break;
3541                         }
3542
3543                         /*
3544                          * workaround: re-program speed mode bit after
3545                          * link-up event
3546                          */
3547                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3548                             !txb2b) {
3549                                 u32 tarc0;
3550                                 tarc0 = er32(TARC(0));
3551                                 tarc0 &= ~SPEED_MODE_BIT;
3552                                 ew32(TARC(0), tarc0);
3553                         }
3554
3555                         /*
3556                          * disable TSO for pcie and 10/100 speeds, to avoid
3557                          * some hardware issues
3558                          */
3559                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
3560                                 switch (adapter->link_speed) {
3561                                 case SPEED_10:
3562                                 case SPEED_100:
3563                                         e_info("10/100 speed: disabling TSO\n");
3564                                         netdev->features &= ~NETIF_F_TSO;
3565                                         netdev->features &= ~NETIF_F_TSO6;
3566                                         break;
3567                                 case SPEED_1000:
3568                                         netdev->features |= NETIF_F_TSO;
3569                                         netdev->features |= NETIF_F_TSO6;
3570                                         break;
3571                                 default:
3572                                         /* oops */
3573                                         break;
3574                                 }
3575                         }
3576
3577                         /*
3578                          * enable transmits in the hardware, need to do this
3579                          * after setting TARC(0)
3580                          */
3581                         tctl = er32(TCTL);
3582                         tctl |= E1000_TCTL_EN;
3583                         ew32(TCTL, tctl);
3584
3585                         netif_carrier_on(netdev);
3586                         netif_tx_wake_all_queues(netdev);
3587
3588                         if (!test_bit(__E1000_DOWN, &adapter->state))
3589                                 mod_timer(&adapter->phy_info_timer,
3590                                           round_jiffies(jiffies + 2 * HZ));
3591                 }
3592         } else {
3593                 if (netif_carrier_ok(netdev)) {
3594                         adapter->link_speed = 0;
3595                         adapter->link_duplex = 0;
3596                         e_info("Link is Down\n");
3597                         netif_carrier_off(netdev);
3598                         netif_tx_stop_all_queues(netdev);
3599                         if (!test_bit(__E1000_DOWN, &adapter->state))
3600                                 mod_timer(&adapter->phy_info_timer,
3601                                           round_jiffies(jiffies + 2 * HZ));
3602
3603                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
3604                                 schedule_work(&adapter->reset_task);
3605                 }
3606         }
3607
3608 link_up:
3609         e1000e_update_stats(adapter);
3610
3611         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
3612         adapter->tpt_old = adapter->stats.tpt;
3613         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
3614         adapter->colc_old = adapter->stats.colc;
3615
3616         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
3617         adapter->gorc_old = adapter->stats.gorc;
3618         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
3619         adapter->gotc_old = adapter->stats.gotc;
3620
3621         e1000e_update_adaptive(&adapter->hw);
3622
3623         if (!netif_carrier_ok(netdev)) {
3624                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
3625                                tx_ring->count);
3626                 if (tx_pending) {
3627                         /*
3628                          * We've lost link, so the controller stops DMA,
3629                          * but we've got queued Tx work that's never going
3630                          * to get done, so reset controller to flush Tx.
3631                          * (Do the reset outside of interrupt context).
3632                          */
3633                         adapter->tx_timeout_count++;
3634                         schedule_work(&adapter->reset_task);
3635                 }
3636         }
3637
3638         /* Cause software interrupt to ensure Rx ring is cleaned */
3639         if (adapter->msix_entries)
3640                 ew32(ICS, adapter->rx_ring->ims_val);
3641         else
3642                 ew32(ICS, E1000_ICS_RXDMT0);
3643
3644         /* Force detection of hung controller every watchdog period */
3645         adapter->detect_tx_hung = 1;
3646
3647         /*
3648          * With 82571 controllers, LAA may be overwritten due to controller
3649          * reset from the other port. Set the appropriate LAA in RAR[0]
3650          */
3651         if (e1000e_get_laa_state_82571(hw))
3652                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
3653
3654         /* Reset the timer */
3655         if (!test_bit(__E1000_DOWN, &adapter->state))
3656                 mod_timer(&adapter->watchdog_timer,
3657                           round_jiffies(jiffies + 2 * HZ));
3658 }
3659
3660 #define E1000_TX_FLAGS_CSUM             0x00000001
3661 #define E1000_TX_FLAGS_VLAN             0x00000002
3662 #define E1000_TX_FLAGS_TSO              0x00000004
3663 #define E1000_TX_FLAGS_IPV4             0x00000008
3664 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
3665 #define E1000_TX_FLAGS_VLAN_SHIFT       16
3666
3667 static int e1000_tso(struct e1000_adapter *adapter,
3668                      struct sk_buff *skb)
3669 {
3670         struct e1000_ring *tx_ring = adapter->tx_ring;
3671         struct e1000_context_desc *context_desc;
3672         struct e1000_buffer *buffer_info;
3673         unsigned int i;
3674         u32 cmd_length = 0;
3675         u16 ipcse = 0, tucse, mss;
3676         u8 ipcss, ipcso, tucss, tucso, hdr_len;
3677         int err;
3678
3679         if (skb_is_gso(skb)) {
3680                 if (skb_header_cloned(skb)) {
3681                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3682                         if (err)
3683                                 return err;
3684                 }
3685
3686                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3687                 mss = skb_shinfo(skb)->gso_size;
3688                 if (skb->protocol == htons(ETH_P_IP)) {
3689                         struct iphdr *iph = ip_hdr(skb);
3690                         iph->tot_len = 0;
3691                         iph->check = 0;
3692                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3693                                                                  iph->daddr, 0,
3694                                                                  IPPROTO_TCP,
3695                                                                  0);
3696                         cmd_length = E1000_TXD_CMD_IP;
3697                         ipcse = skb_transport_offset(skb) - 1;
3698                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3699                         ipv6_hdr(skb)->payload_len = 0;
3700                         tcp_hdr(skb)->check =
3701                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3702                                                  &ipv6_hdr(skb)->daddr,
3703                                                  0, IPPROTO_TCP, 0);
3704                         ipcse = 0;
3705                 }
3706                 ipcss = skb_network_offset(skb);
3707                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
3708                 tucss = skb_transport_offset(skb);
3709                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
3710                 tucse = 0;
3711
3712                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
3713                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
3714
3715                 i = tx_ring->next_to_use;
3716                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3717                 buffer_info = &tx_ring->buffer_info[i];
3718
3719                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
3720                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
3721                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
3722                 context_desc->upper_setup.tcp_fields.tucss = tucss;
3723                 context_desc->upper_setup.tcp_fields.tucso = tucso;
3724                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
3725                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
3726                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
3727                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
3728
3729                 buffer_info->time_stamp = jiffies;
3730                 buffer_info->next_to_watch = i;
3731
3732                 i++;
3733                 if (i == tx_ring->count)
3734                         i = 0;
3735                 tx_ring->next_to_use = i;
3736
3737                 return 1;
3738         }
3739
3740         return 0;
3741 }
3742
3743 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
3744 {
3745         struct e1000_ring *tx_ring = adapter->tx_ring;
3746         struct e1000_context_desc *context_desc;
3747         struct e1000_buffer *buffer_info;
3748         unsigned int i;
3749         u8 css;
3750         u32 cmd_len = E1000_TXD_CMD_DEXT;
3751
3752         if (skb->ip_summed != CHECKSUM_PARTIAL)
3753                 return 0;
3754
3755         switch (skb->protocol) {
3756         case __constant_htons(ETH_P_IP):
3757                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3758                         cmd_len |= E1000_TXD_CMD_TCP;
3759                 break;
3760         case __constant_htons(ETH_P_IPV6):
3761                 /* XXX not handling all IPV6 headers */
3762                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3763                         cmd_len |= E1000_TXD_CMD_TCP;
3764                 break;
3765         default:
3766                 if (unlikely(net_ratelimit()))
3767                         e_warn("checksum_partial proto=%x!\n", skb->protocol);
3768                 break;
3769         }
3770
3771         css = skb_transport_offset(skb);
3772
3773         i = tx_ring->next_to_use;
3774         buffer_info = &tx_ring->buffer_info[i];
3775         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
3776
3777         context_desc->lower_setup.ip_config = 0;
3778         context_desc->upper_setup.tcp_fields.tucss = css;
3779         context_desc->upper_setup.tcp_fields.tucso =
3780                                 css + skb->csum_offset;
3781         context_desc->upper_setup.tcp_fields.tucse = 0;
3782         context_desc->tcp_seg_setup.data = 0;
3783         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
3784
3785         buffer_info->time_stamp = jiffies;
3786         buffer_info->next_to_watch = i;
3787
3788         i++;
3789         if (i == tx_ring->count)
3790                 i = 0;
3791         tx_ring->next_to_use = i;
3792
3793         return 1;
3794 }
3795
3796 #define E1000_MAX_PER_TXD       8192
3797 #define E1000_MAX_TXD_PWR       12
3798
3799 static int e1000_tx_map(struct e1000_adapter *adapter,
3800                         struct sk_buff *skb, unsigned int first,
3801                         unsigned int max_per_txd, unsigned int nr_frags,
3802                         unsigned int mss)
3803 {
3804         struct e1000_ring *tx_ring = adapter->tx_ring;
3805         struct e1000_buffer *buffer_info;
3806         unsigned int len = skb->len - skb->data_len;
3807         unsigned int offset = 0, size, count = 0, i;
3808         unsigned int f;
3809
3810         i = tx_ring->next_to_use;
3811
3812         while (len) {
3813                 buffer_info = &tx_ring->buffer_info[i];
3814                 size = min(len, max_per_txd);
3815
3816                 /* Workaround for premature desc write-backs
3817                  * in TSO mode.  Append 4-byte sentinel desc */
3818                 if (mss && !nr_frags && size == len && size > 8)
3819                         size -= 4;
3820
3821                 buffer_info->length = size;
3822                 /* set time_stamp *before* dma to help avoid a possible race */
3823                 buffer_info->time_stamp = jiffies;
3824                 buffer_info->dma =
3825                         pci_map_single(adapter->pdev,
3826                                 skb->data + offset,
3827                                 size,
3828                                 PCI_DMA_TODEVICE);
3829                 if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
3830                         dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3831                         adapter->tx_dma_failed++;
3832                         return -1;
3833                 }
3834                 buffer_info->next_to_watch = i;
3835
3836                 len -= size;
3837                 offset += size;
3838                 count++;
3839                 i++;
3840                 if (i == tx_ring->count)
3841                         i = 0;
3842         }
3843
3844         for (f = 0; f < nr_frags; f++) {
3845                 struct skb_frag_struct *frag;
3846
3847                 frag = &skb_shinfo(skb)->frags[f];
3848                 len = frag->size;
3849                 offset = frag->page_offset;
3850
3851                 while (len) {
3852                         buffer_info = &tx_ring->buffer_info[i];
3853                         size = min(len, max_per_txd);
3854                         /* Workaround for premature desc write-backs
3855                          * in TSO mode.  Append 4-byte sentinel desc */
3856                         if (mss && f == (nr_frags-1) && size == len && size > 8)
3857                                 size -= 4;
3858
3859                         buffer_info->length = size;
3860                         buffer_info->time_stamp = jiffies;
3861                         buffer_info->dma =
3862                                 pci_map_page(adapter->pdev,
3863                                         frag->page,
3864                                         offset,
3865                                         size,
3866                                         PCI_DMA_TODEVICE);
3867                         if (pci_dma_mapping_error(adapter->pdev,
3868                                                   buffer_info->dma)) {
3869                                 dev_err(&adapter->pdev->dev,
3870                                         "TX DMA page map failed\n");
3871                                 adapter->tx_dma_failed++;
3872                                 return -1;
3873                         }
3874
3875                         buffer_info->next_to_watch = i;
3876
3877                         len -= size;
3878                         offset += size;
3879                         count++;
3880
3881                         i++;
3882                         if (i == tx_ring->count)
3883                                 i = 0;
3884                 }
3885         }
3886
3887         if (i == 0)
3888                 i = tx_ring->count - 1;
3889         else
3890                 i--;
3891
3892         tx_ring->buffer_info[i].skb = skb;
3893         tx_ring->buffer_info[first].next_to_watch = i;
3894
3895         return count;
3896 }
3897
3898 static void e1000_tx_queue(struct e1000_adapter *adapter,
3899                            int tx_flags, int count)
3900 {
3901         struct e1000_ring *tx_ring = adapter->tx_ring;
3902         struct e1000_tx_desc *tx_desc = NULL;
3903         struct e1000_buffer *buffer_info;
3904         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
3905         unsigned int i;
3906
3907         if (tx_flags & E1000_TX_FLAGS_TSO) {
3908                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
3909                              E1000_TXD_CMD_TSE;
3910                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3911
3912                 if (tx_flags & E1000_TX_FLAGS_IPV4)
3913                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
3914         }
3915
3916         if (tx_flags & E1000_TX_FLAGS_CSUM) {
3917                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3918                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3919         }
3920
3921         if (tx_flags & E1000_TX_FLAGS_VLAN) {
3922                 txd_lower |= E1000_TXD_CMD_VLE;
3923                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3924         }
3925
3926         i = tx_ring->next_to_use;
3927
3928         while (count--) {
3929                 buffer_info = &tx_ring->buffer_info[i];
3930                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3931                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3932                 tx_desc->lower.data =
3933                         cpu_to_le32(txd_lower | buffer_info->length);
3934                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3935
3936                 i++;
3937                 if (i == tx_ring->count)
3938                         i = 0;
3939         }
3940
3941         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3942
3943         /*
3944          * Force memory writes to complete before letting h/w
3945          * know there are new descriptors to fetch.  (Only
3946          * applicable for weak-ordered memory model archs,
3947          * such as IA-64).
3948          */
3949         wmb();
3950
3951         tx_ring->next_to_use = i;
3952         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3953         /*
3954          * we need this if more than one processor can write to our tail
3955          * at a time, it synchronizes IO on IA64/Altix systems
3956          */
3957         mmiowb();
3958 }
3959
3960 #define MINIMUM_DHCP_PACKET_SIZE 282
3961 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
3962                                     struct sk_buff *skb)
3963 {
3964         struct e1000_hw *hw =  &adapter->hw;
3965         u16 length, offset;
3966
3967         if (vlan_tx_tag_present(skb)) {
3968                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id)
3969                     && (adapter->hw.mng_cookie.status &
3970                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
3971                         return 0;
3972         }
3973
3974         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
3975                 return 0;
3976
3977         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
3978                 return 0;
3979
3980         {
3981                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
3982                 struct udphdr *udp;
3983
3984                 if (ip->protocol != IPPROTO_UDP)
3985                         return 0;
3986
3987                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
3988                 if (ntohs(udp->dest) != 67)
3989                         return 0;
3990
3991                 offset = (u8 *)udp + 8 - skb->data;
3992                 length = skb->len - offset;
3993                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
3994         }
3995
3996         return 0;
3997 }
3998
3999 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4000 {
4001         struct e1000_adapter *adapter = netdev_priv(netdev);
4002
4003         netif_stop_queue(netdev);
4004         /*
4005          * Herbert's original patch had:
4006          *  smp_mb__after_netif_stop_queue();
4007          * but since that doesn't exist yet, just open code it.
4008          */
4009         smp_mb();
4010
4011         /*
4012          * We need to check again in a case another CPU has just
4013          * made room available.
4014          */
4015         if (e1000_desc_unused(adapter->tx_ring) < size)
4016                 return -EBUSY;
4017
4018         /* A reprieve! */
4019         netif_start_queue(netdev);
4020         ++adapter->restart_queue;
4021         return 0;
4022 }
4023
4024 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4025 {
4026         struct e1000_adapter *adapter = netdev_priv(netdev);
4027
4028         if (e1000_desc_unused(adapter->tx_ring) >= size)
4029                 return 0;
4030         return __e1000_maybe_stop_tx(netdev, size);
4031 }
4032
4033 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4034 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4035 {
4036         struct e1000_adapter *adapter = netdev_priv(netdev);
4037         struct e1000_ring *tx_ring = adapter->tx_ring;
4038         unsigned int first;
4039         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4040         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4041         unsigned int tx_flags = 0;
4042         unsigned int len = skb->len - skb->data_len;
4043         unsigned long irq_flags;
4044         unsigned int nr_frags;
4045         unsigned int mss;
4046         int count = 0;
4047         int tso;
4048         unsigned int f;
4049
4050         if (test_bit(__E1000_DOWN, &adapter->state)) {
4051                 dev_kfree_skb_any(skb);
4052                 return NETDEV_TX_OK;
4053         }
4054
4055         if (skb->len <= 0) {
4056                 dev_kfree_skb_any(skb);
4057                 return NETDEV_TX_OK;
4058         }
4059
4060         mss = skb_shinfo(skb)->gso_size;
4061         /*
4062          * The controller does a simple calculation to
4063          * make sure there is enough room in the FIFO before
4064          * initiating the DMA for each buffer.  The calc is:
4065          * 4 = ceil(buffer len/mss).  To make sure we don't
4066          * overrun the FIFO, adjust the max buffer len if mss
4067          * drops.
4068          */
4069         if (mss) {
4070                 u8 hdr_len;
4071                 max_per_txd = min(mss << 2, max_per_txd);
4072                 max_txd_pwr = fls(max_per_txd) - 1;
4073
4074                 /*
4075                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4076                  * points to just header, pull a few bytes of payload from
4077                  * frags into skb->data
4078                  */
4079                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4080                 /*
4081                  * we do this workaround for ES2LAN, but it is un-necessary,
4082                  * avoiding it could save a lot of cycles
4083                  */
4084                 if (skb->data_len && (hdr_len == len)) {
4085                         unsigned int pull_size;
4086
4087                         pull_size = min((unsigned int)4, skb->data_len);
4088                         if (!__pskb_pull_tail(skb, pull_size)) {
4089                                 e_err("__pskb_pull_tail failed.\n");
4090                                 dev_kfree_skb_any(skb);
4091                                 return NETDEV_TX_OK;
4092                         }
4093                         len = skb->len - skb->data_len;
4094                 }
4095         }
4096
4097         /* reserve a descriptor for the offload context */
4098         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4099                 count++;
4100         count++;
4101
4102         count += TXD_USE_COUNT(len, max_txd_pwr);
4103
4104         nr_frags = skb_shinfo(skb)->nr_frags;
4105         for (f = 0; f < nr_frags; f++)
4106                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4107                                        max_txd_pwr);
4108
4109         if (adapter->hw.mac.tx_pkt_filtering)
4110                 e1000_transfer_dhcp_info(adapter, skb);
4111
4112         if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
4113                 /* Collision - tell upper layer to requeue */
4114                 return NETDEV_TX_LOCKED;
4115
4116         /*
4117          * need: count + 2 desc gap to keep tail from touching
4118          * head, otherwise try next time
4119          */
4120         if (e1000_maybe_stop_tx(netdev, count + 2)) {
4121                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4122                 return NETDEV_TX_BUSY;
4123         }
4124
4125         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4126                 tx_flags |= E1000_TX_FLAGS_VLAN;
4127                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4128         }
4129
4130         first = tx_ring->next_to_use;
4131
4132         tso = e1000_tso(adapter, skb);
4133         if (tso < 0) {
4134                 dev_kfree_skb_any(skb);
4135                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4136                 return NETDEV_TX_OK;
4137         }
4138
4139         if (tso)
4140                 tx_flags |= E1000_TX_FLAGS_TSO;
4141         else if (e1000_tx_csum(adapter, skb))
4142                 tx_flags |= E1000_TX_FLAGS_CSUM;
4143
4144         /*
4145          * Old method was to assume IPv4 packet by default if TSO was enabled.
4146          * 82571 hardware supports TSO capabilities for IPv6 as well...
4147          * no longer assume, we must.
4148          */
4149         if (skb->protocol == htons(ETH_P_IP))
4150                 tx_flags |= E1000_TX_FLAGS_IPV4;
4151
4152         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4153         if (count < 0) {
4154                 /* handle pci_map_single() error in e1000_tx_map */
4155                 dev_kfree_skb_any(skb);
4156                 spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4157                 return NETDEV_TX_OK;
4158         }
4159
4160         e1000_tx_queue(adapter, tx_flags, count);
4161
4162         netdev->trans_start = jiffies;
4163
4164         /* Make sure there is space in the ring for the next send. */
4165         e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4166
4167         spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
4168         return NETDEV_TX_OK;
4169 }
4170
4171 /**
4172  * e1000_tx_timeout - Respond to a Tx Hang
4173  * @netdev: network interface device structure
4174  **/
4175 static void e1000_tx_timeout(struct net_device *netdev)
4176 {
4177         struct e1000_adapter *adapter = netdev_priv(netdev);
4178
4179         /* Do the reset outside of interrupt context */
4180         adapter->tx_timeout_count++;
4181         schedule_work(&adapter->reset_task);
4182 }
4183
4184 static void e1000_reset_task(struct work_struct *work)
4185 {
4186         struct e1000_adapter *adapter;
4187         adapter = container_of(work, struct e1000_adapter, reset_task);
4188
4189         e1000e_reinit_locked(adapter);
4190 }
4191
4192 /**
4193  * e1000_get_stats - Get System Network Statistics
4194  * @netdev: network interface device structure
4195  *
4196  * Returns the address of the device statistics structure.
4197  * The statistics are actually updated from the timer callback.
4198  **/
4199 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4200 {
4201         struct e1000_adapter *adapter = netdev_priv(netdev);
4202
4203         /* only return the current stats */
4204         return &adapter->net_stats;
4205 }
4206
4207 /**
4208  * e1000_change_mtu - Change the Maximum Transfer Unit
4209  * @netdev: network interface device structure
4210  * @new_mtu: new value for maximum frame size
4211  *
4212  * Returns 0 on success, negative on failure
4213  **/
4214 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4215 {
4216         struct e1000_adapter *adapter = netdev_priv(netdev);
4217         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4218
4219         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4220             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4221                 e_err("Invalid MTU setting\n");
4222                 return -EINVAL;
4223         }
4224
4225         /* Jumbo frame size limits */
4226         if (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) {
4227                 if (!(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4228                         e_err("Jumbo Frames not supported.\n");
4229                         return -EINVAL;
4230                 }
4231                 if (adapter->hw.phy.type == e1000_phy_ife) {
4232                         e_err("Jumbo Frames not supported.\n");
4233                         return -EINVAL;
4234                 }
4235         }
4236
4237 #define MAX_STD_JUMBO_FRAME_SIZE 9234
4238         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4239                 e_err("MTU > 9216 not supported.\n");
4240                 return -EINVAL;
4241         }
4242
4243         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4244                 msleep(1);
4245         /* e1000e_down has a dependency on max_frame_size */
4246         adapter->max_frame_size = max_frame;
4247         if (netif_running(netdev))
4248                 e1000e_down(adapter);
4249
4250         /*
4251          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4252          * means we reserve 2 more, this pushes us to allocate from the next
4253          * larger slab size.
4254          * i.e. RXBUFFER_2048 --> size-4096 slab
4255          * However with the new *_jumbo_rx* routines, jumbo receives will use
4256          * fragmented skbs
4257          */
4258
4259         if (max_frame <= 256)
4260                 adapter->rx_buffer_len = 256;
4261         else if (max_frame <= 512)
4262                 adapter->rx_buffer_len = 512;
4263         else if (max_frame <= 1024)
4264                 adapter->rx_buffer_len = 1024;
4265         else if (max_frame <= 2048)
4266                 adapter->rx_buffer_len = 2048;
4267         else
4268                 adapter->rx_buffer_len = 4096;
4269
4270         /* adjust allocation if LPE protects us, and we aren't using SBP */
4271         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4272              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4273                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4274                                          + ETH_FCS_LEN;
4275
4276         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4277         netdev->mtu = new_mtu;
4278
4279         if (netif_running(netdev))
4280                 e1000e_up(adapter);
4281         else
4282                 e1000e_reset(adapter);
4283
4284         clear_bit(__E1000_RESETTING, &adapter->state);
4285
4286         return 0;
4287 }
4288
4289 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4290                            int cmd)
4291 {
4292         struct e1000_adapter *adapter = netdev_priv(netdev);
4293         struct mii_ioctl_data *data = if_mii(ifr);
4294
4295         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4296                 return -EOPNOTSUPP;
4297
4298         switch (cmd) {
4299         case SIOCGMIIPHY:
4300                 data->phy_id = adapter->hw.phy.addr;
4301                 break;
4302         case SIOCGMIIREG:
4303                 if (!capable(CAP_NET_ADMIN))
4304                         return -EPERM;
4305                 switch (data->reg_num & 0x1F) {
4306                 case MII_BMCR:
4307                         data->val_out = adapter->phy_regs.bmcr;
4308                         break;
4309                 case MII_BMSR:
4310                         data->val_out = adapter->phy_regs.bmsr;
4311                         break;
4312                 case MII_PHYSID1:
4313                         data->val_out = (adapter->hw.phy.id >> 16);
4314                         break;
4315                 case MII_PHYSID2:
4316                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4317                         break;
4318                 case MII_ADVERTISE:
4319                         data->val_out = adapter->phy_regs.advertise;
4320                         break;
4321                 case MII_LPA:
4322                         data->val_out = adapter->phy_regs.lpa;
4323                         break;
4324                 case MII_EXPANSION:
4325                         data->val_out = adapter->phy_regs.expansion;
4326                         break;
4327                 case MII_CTRL1000:
4328                         data->val_out = adapter->phy_regs.ctrl1000;
4329                         break;
4330                 case MII_STAT1000:
4331                         data->val_out = adapter->phy_regs.stat1000;
4332                         break;
4333                 case MII_ESTATUS:
4334                         data->val_out = adapter->phy_regs.estatus;
4335                         break;
4336                 default:
4337                         return -EIO;
4338                 }
4339                 break;
4340         case SIOCSMIIREG:
4341         default:
4342                 return -EOPNOTSUPP;
4343         }
4344         return 0;
4345 }
4346
4347 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4348 {
4349         switch (cmd) {
4350         case SIOCGMIIPHY:
4351         case SIOCGMIIREG:
4352         case SIOCSMIIREG:
4353                 return e1000_mii_ioctl(netdev, ifr, cmd);
4354         default:
4355                 return -EOPNOTSUPP;
4356         }
4357 }
4358
4359 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4360 {
4361         struct net_device *netdev = pci_get_drvdata(pdev);
4362         struct e1000_adapter *adapter = netdev_priv(netdev);
4363         struct e1000_hw *hw = &adapter->hw;
4364         u32 ctrl, ctrl_ext, rctl, status;
4365         u32 wufc = adapter->wol;
4366         int retval = 0;
4367
4368         netif_device_detach(netdev);
4369
4370         if (netif_running(netdev)) {
4371                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4372                 e1000e_down(adapter);
4373                 e1000_free_irq(adapter);
4374         }
4375         e1000e_reset_interrupt_capability(adapter);
4376
4377         retval = pci_save_state(pdev);
4378         if (retval)
4379                 return retval;
4380
4381         status = er32(STATUS);
4382         if (status & E1000_STATUS_LU)
4383                 wufc &= ~E1000_WUFC_LNKC;
4384
4385         if (wufc) {
4386                 e1000_setup_rctl(adapter);
4387                 e1000_set_multi(netdev);
4388
4389                 /* turn on all-multi mode if wake on multicast is enabled */
4390                 if (wufc & E1000_WUFC_MC) {
4391                         rctl = er32(RCTL);
4392                         rctl |= E1000_RCTL_MPE;
4393                         ew32(RCTL, rctl);
4394                 }
4395
4396                 ctrl = er32(CTRL);
4397                 /* advertise wake from D3Cold */
4398                 #define E1000_CTRL_ADVD3WUC 0x00100000
4399                 /* phy power management enable */
4400                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4401                 ctrl |= E1000_CTRL_ADVD3WUC |
4402                         E1000_CTRL_EN_PHY_PWR_MGMT;
4403                 ew32(CTRL, ctrl);
4404
4405                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4406                     adapter->hw.phy.media_type ==
4407                     e1000_media_type_internal_serdes) {
4408                         /* keep the laser running in D3 */
4409                         ctrl_ext = er32(CTRL_EXT);
4410                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4411                         ew32(CTRL_EXT, ctrl_ext);
4412                 }
4413
4414                 if (adapter->flags & FLAG_IS_ICH)
4415                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4416
4417                 /* Allow time for pending master requests to run */
4418                 e1000e_disable_pcie_master(&adapter->hw);
4419
4420                 ew32(WUC, E1000_WUC_PME_EN);
4421                 ew32(WUFC, wufc);
4422                 pci_enable_wake(pdev, PCI_D3hot, 1);
4423                 pci_enable_wake(pdev, PCI_D3cold, 1);
4424         } else {
4425                 ew32(WUC, 0);
4426                 ew32(WUFC, 0);
4427                 pci_enable_wake(pdev, PCI_D3hot, 0);
4428                 pci_enable_wake(pdev, PCI_D3cold, 0);
4429         }
4430
4431         /* make sure adapter isn't asleep if manageability is enabled */
4432         if (adapter->flags & FLAG_MNG_PT_ENABLED) {
4433                 pci_enable_wake(pdev, PCI_D3hot, 1);
4434                 pci_enable_wake(pdev, PCI_D3cold, 1);
4435         }
4436
4437         if (adapter->hw.phy.type == e1000_phy_igp_3)
4438                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4439
4440         /*
4441          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4442          * would have already happened in close and is redundant.
4443          */
4444         e1000_release_hw_control(adapter);
4445
4446         pci_disable_device(pdev);
4447
4448         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4449
4450         return 0;
4451 }
4452
4453 static void e1000e_disable_l1aspm(struct pci_dev *pdev)
4454 {
4455         int pos;
4456         u16 val;
4457
4458         /*
4459          * 82573 workaround - disable L1 ASPM on mobile chipsets
4460          *
4461          * L1 ASPM on various mobile (ich7) chipsets do not behave properly
4462          * resulting in lost data or garbage information on the pci-e link
4463          * level. This could result in (false) bad EEPROM checksum errors,
4464          * long ping times (up to 2s) or even a system freeze/hang.
4465          *
4466          * Unfortunately this feature saves about 1W power consumption when
4467          * active.
4468          */
4469         pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
4470         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &val);
4471         if (val & 0x2) {
4472                 dev_warn(&pdev->dev, "Disabling L1 ASPM\n");
4473                 val &= ~0x2;
4474                 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, val);
4475         }
4476 }
4477
4478 #ifdef CONFIG_PM
4479 static int e1000_resume(struct pci_dev *pdev)
4480 {
4481         struct net_device *netdev = pci_get_drvdata(pdev);
4482         struct e1000_adapter *adapter = netdev_priv(netdev);
4483         struct e1000_hw *hw = &adapter->hw;
4484         u32 err;
4485
4486         pci_set_power_state(pdev, PCI_D0);
4487         pci_restore_state(pdev);
4488         e1000e_disable_l1aspm(pdev);
4489
4490         err = pci_enable_device_mem(pdev);
4491         if (err) {
4492                 dev_err(&pdev->dev,
4493                         "Cannot enable PCI device from suspend\n");
4494                 return err;
4495         }
4496
4497         pci_set_master(pdev);
4498
4499         pci_enable_wake(pdev, PCI_D3hot, 0);
4500         pci_enable_wake(pdev, PCI_D3cold, 0);
4501
4502         e1000e_set_interrupt_capability(adapter);
4503         if (netif_running(netdev)) {
4504                 err = e1000_request_irq(adapter);
4505                 if (err)
4506                         return err;
4507         }
4508
4509         e1000e_power_up_phy(adapter);
4510         e1000e_reset(adapter);
4511         ew32(WUS, ~0);
4512
4513         e1000_init_manageability(adapter);
4514
4515         if (netif_running(netdev))
4516                 e1000e_up(adapter);
4517
4518         netif_device_attach(netdev);
4519
4520         /*
4521          * If the controller has AMT, do not set DRV_LOAD until the interface
4522          * is up.  For all other cases, let the f/w know that the h/w is now
4523          * under the control of the driver.
4524          */
4525         if (!(adapter->flags & FLAG_HAS_AMT))
4526                 e1000_get_hw_control(adapter);
4527
4528         return 0;
4529 }
4530 #endif
4531
4532 static void e1000_shutdown(struct pci_dev *pdev)
4533 {
4534         e1000_suspend(pdev, PMSG_SUSPEND);
4535 }
4536
4537 #ifdef CONFIG_NET_POLL_CONTROLLER
4538 /*
4539  * Polling 'interrupt' - used by things like netconsole to send skbs
4540  * without having to re-enable interrupts. It's not called while
4541  * the interrupt routine is executing.
4542  */
4543 static void e1000_netpoll(struct net_device *netdev)
4544 {
4545         struct e1000_adapter *adapter = netdev_priv(netdev);
4546
4547         disable_irq(adapter->pdev->irq);
4548         e1000_intr(adapter->pdev->irq, netdev);
4549
4550         enable_irq(adapter->pdev->irq);
4551 }
4552 #endif
4553
4554 /**
4555  * e1000_io_error_detected - called when PCI error is detected
4556  * @pdev: Pointer to PCI device
4557  * @state: The current pci connection state
4558  *
4559  * This function is called after a PCI bus error affecting
4560  * this device has been detected.
4561  */
4562 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4563                                                 pci_channel_state_t state)
4564 {
4565         struct net_device *netdev = pci_get_drvdata(pdev);
4566         struct e1000_adapter *adapter = netdev_priv(netdev);
4567
4568         netif_device_detach(netdev);
4569
4570         if (netif_running(netdev))
4571                 e1000e_down(adapter);
4572         pci_disable_device(pdev);
4573
4574         /* Request a slot slot reset. */
4575         return PCI_ERS_RESULT_NEED_RESET;
4576 }
4577
4578 /**
4579  * e1000_io_slot_reset - called after the pci bus has been reset.
4580  * @pdev: Pointer to PCI device
4581  *
4582  * Restart the card from scratch, as if from a cold-boot. Implementation
4583  * resembles the first-half of the e1000_resume routine.
4584  */
4585 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4586 {
4587         struct net_device *netdev = pci_get_drvdata(pdev);
4588         struct e1000_adapter *adapter = netdev_priv(netdev);
4589         struct e1000_hw *hw = &adapter->hw;
4590         int err;
4591
4592         e1000e_disable_l1aspm(pdev);
4593         err = pci_enable_device_mem(pdev);
4594         if (err) {
4595                 dev_err(&pdev->dev,
4596                         "Cannot re-enable PCI device after reset.\n");
4597                 return PCI_ERS_RESULT_DISCONNECT;
4598         }
4599         pci_set_master(pdev);
4600         pci_restore_state(pdev);
4601
4602         pci_enable_wake(pdev, PCI_D3hot, 0);
4603         pci_enable_wake(pdev, PCI_D3cold, 0);
4604
4605         e1000e_reset(adapter);
4606         ew32(WUS, ~0);
4607
4608         return PCI_ERS_RESULT_RECOVERED;
4609 }
4610
4611 /**
4612  * e1000_io_resume - called when traffic can start flowing again.
4613  * @pdev: Pointer to PCI device
4614  *
4615  * This callback is called when the error recovery driver tells us that
4616  * its OK to resume normal operation. Implementation resembles the
4617  * second-half of the e1000_resume routine.
4618  */
4619 static void e1000_io_resume(struct pci_dev *pdev)
4620 {
4621         struct net_device *netdev = pci_get_drvdata(pdev);
4622         struct e1000_adapter *adapter = netdev_priv(netdev);
4623
4624         e1000_init_manageability(adapter);
4625
4626         if (netif_running(netdev)) {
4627                 if (e1000e_up(adapter)) {
4628                         dev_err(&pdev->dev,
4629                                 "can't bring device back up after reset\n");
4630                         return;
4631                 }
4632         }
4633
4634         netif_device_attach(netdev);
4635
4636         /*
4637          * If the controller has AMT, do not set DRV_LOAD until the interface
4638          * is up.  For all other cases, let the f/w know that the h/w is now
4639          * under the control of the driver.
4640          */
4641         if (!(adapter->flags & FLAG_HAS_AMT))
4642                 e1000_get_hw_control(adapter);
4643
4644 }
4645
4646 static void e1000_print_device_info(struct e1000_adapter *adapter)
4647 {
4648         struct e1000_hw *hw = &adapter->hw;
4649         struct net_device *netdev = adapter->netdev;
4650         u32 pba_num;
4651
4652         /* print bus type/speed/width info */
4653         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
4654                /* bus width */
4655                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
4656                 "Width x1"),
4657                /* MAC address */
4658                netdev->dev_addr);
4659         e_info("Intel(R) PRO/%s Network Connection\n",
4660                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
4661         e1000e_read_pba_num(hw, &pba_num);
4662         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4663                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
4664 }
4665
4666 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
4667 {
4668         struct e1000_hw *hw = &adapter->hw;
4669         int ret_val;
4670         u16 buf = 0;
4671
4672         if (hw->mac.type != e1000_82573)
4673                 return;
4674
4675         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
4676         if (!(le16_to_cpu(buf) & (1 << 0))) {
4677                 /* Deep Smart Power Down (DSPD) */
4678                 dev_warn(&adapter->pdev->dev,
4679                          "Warning: detected DSPD enabled in EEPROM\n");
4680         }
4681
4682         ret_val = e1000_read_nvm(hw, NVM_INIT_3GIO_3, 1, &buf);
4683         if (le16_to_cpu(buf) & (3 << 2)) {
4684                 /* ASPM enable */
4685                 dev_warn(&adapter->pdev->dev,
4686                          "Warning: detected ASPM enabled in EEPROM\n");
4687         }
4688 }
4689
4690 /**
4691  * e1000_probe - Device Initialization Routine
4692  * @pdev: PCI device information struct
4693  * @ent: entry in e1000_pci_tbl
4694  *
4695  * Returns 0 on success, negative on failure
4696  *
4697  * e1000_probe initializes an adapter identified by a pci_dev structure.
4698  * The OS initialization, configuring of the adapter private structure,
4699  * and a hardware reset occur.
4700  **/
4701 static int __devinit e1000_probe(struct pci_dev *pdev,
4702                                  const struct pci_device_id *ent)
4703 {
4704         struct net_device *netdev;
4705         struct e1000_adapter *adapter;
4706         struct e1000_hw *hw;
4707         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
4708         resource_size_t mmio_start, mmio_len;
4709         resource_size_t flash_start, flash_len;
4710
4711         static int cards_found;
4712         int i, err, pci_using_dac;
4713         u16 eeprom_data = 0;
4714         u16 eeprom_apme_mask = E1000_EEPROM_APME;
4715
4716         e1000e_disable_l1aspm(pdev);
4717
4718         err = pci_enable_device_mem(pdev);
4719         if (err)
4720                 return err;
4721
4722         pci_using_dac = 0;
4723         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
4724         if (!err) {
4725                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
4726                 if (!err)
4727                         pci_using_dac = 1;
4728         } else {
4729                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4730                 if (err) {
4731                         err = pci_set_consistent_dma_mask(pdev,
4732                                                           DMA_32BIT_MASK);
4733                         if (err) {
4734                                 dev_err(&pdev->dev, "No usable DMA "
4735                                         "configuration, aborting\n");
4736                                 goto err_dma;
4737                         }
4738                 }
4739         }
4740
4741         err = pci_request_selected_regions(pdev,
4742                                           pci_select_bars(pdev, IORESOURCE_MEM),
4743                                           e1000e_driver_name);
4744         if (err)
4745                 goto err_pci_reg;
4746
4747         pci_set_master(pdev);
4748         pci_save_state(pdev);
4749
4750         err = -ENOMEM;
4751         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
4752         if (!netdev)
4753                 goto err_alloc_etherdev;
4754
4755         SET_NETDEV_DEV(netdev, &pdev->dev);
4756
4757         pci_set_drvdata(pdev, netdev);
4758         adapter = netdev_priv(netdev);
4759         hw = &adapter->hw;
4760         adapter->netdev = netdev;
4761         adapter->pdev = pdev;
4762         adapter->ei = ei;
4763         adapter->pba = ei->pba;
4764         adapter->flags = ei->flags;
4765         adapter->hw.adapter = adapter;
4766         adapter->hw.mac.type = ei->mac;
4767         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
4768
4769         mmio_start = pci_resource_start(pdev, 0);
4770         mmio_len = pci_resource_len(pdev, 0);
4771
4772         err = -EIO;
4773         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
4774         if (!adapter->hw.hw_addr)
4775                 goto err_ioremap;
4776
4777         if ((adapter->flags & FLAG_HAS_FLASH) &&
4778             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
4779                 flash_start = pci_resource_start(pdev, 1);
4780                 flash_len = pci_resource_len(pdev, 1);
4781                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
4782                 if (!adapter->hw.flash_address)
4783                         goto err_flashmap;
4784         }
4785
4786         /* construct the net_device struct */
4787         netdev->open                    = &e1000_open;
4788         netdev->stop                    = &e1000_close;
4789         netdev->hard_start_xmit         = &e1000_xmit_frame;
4790         netdev->get_stats               = &e1000_get_stats;
4791         netdev->set_multicast_list      = &e1000_set_multi;
4792         netdev->set_mac_address         = &e1000_set_mac;
4793         netdev->change_mtu              = &e1000_change_mtu;
4794         netdev->do_ioctl                = &e1000_ioctl;
4795         e1000e_set_ethtool_ops(netdev);
4796         netdev->tx_timeout              = &e1000_tx_timeout;
4797         netdev->watchdog_timeo          = 5 * HZ;
4798         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
4799         netdev->vlan_rx_register        = e1000_vlan_rx_register;
4800         netdev->vlan_rx_add_vid         = e1000_vlan_rx_add_vid;
4801         netdev->vlan_rx_kill_vid        = e1000_vlan_rx_kill_vid;
4802 #ifdef CONFIG_NET_POLL_CONTROLLER
4803         netdev->poll_controller         = e1000_netpoll;
4804 #endif
4805         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
4806
4807         netdev->mem_start = mmio_start;
4808         netdev->mem_end = mmio_start + mmio_len;
4809
4810         adapter->bd_number = cards_found++;
4811
4812         e1000e_check_options(adapter);
4813
4814         /* setup adapter struct */
4815         err = e1000_sw_init(adapter);
4816         if (err)
4817                 goto err_sw_init;
4818
4819         err = -EIO;
4820
4821         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4822         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
4823         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4824
4825         err = ei->get_variants(adapter);
4826         if (err)
4827                 goto err_hw_init;
4828
4829         if ((adapter->flags & FLAG_IS_ICH) &&
4830             (adapter->flags & FLAG_READ_ONLY_NVM))
4831                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
4832
4833         hw->mac.ops.get_bus_info(&adapter->hw);
4834
4835         adapter->hw.phy.autoneg_wait_to_complete = 0;
4836
4837         /* Copper options */
4838         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
4839                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
4840                 adapter->hw.phy.disable_polarity_correction = 0;
4841                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
4842         }
4843
4844         if (e1000_check_reset_block(&adapter->hw))
4845                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
4846
4847         netdev->features = NETIF_F_SG |
4848                            NETIF_F_HW_CSUM |
4849                            NETIF_F_HW_VLAN_TX |
4850                            NETIF_F_HW_VLAN_RX;
4851
4852         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
4853                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
4854
4855         netdev->features |= NETIF_F_TSO;
4856         netdev->features |= NETIF_F_TSO6;
4857
4858         netdev->vlan_features |= NETIF_F_TSO;
4859         netdev->vlan_features |= NETIF_F_TSO6;
4860         netdev->vlan_features |= NETIF_F_HW_CSUM;
4861         netdev->vlan_features |= NETIF_F_SG;
4862
4863         if (pci_using_dac)
4864                 netdev->features |= NETIF_F_HIGHDMA;
4865
4866         /*
4867          * We should not be using LLTX anymore, but we are still Tx faster with
4868          * it.
4869          */
4870         netdev->features |= NETIF_F_LLTX;
4871
4872         if (e1000e_enable_mng_pass_thru(&adapter->hw))
4873                 adapter->flags |= FLAG_MNG_PT_ENABLED;
4874
4875         /*
4876          * before reading the NVM, reset the controller to
4877          * put the device in a known good starting state
4878          */
4879         adapter->hw.mac.ops.reset_hw(&adapter->hw);
4880
4881         /*
4882          * systems with ASPM and others may see the checksum fail on the first
4883          * attempt. Let's give it a few tries
4884          */
4885         for (i = 0;; i++) {
4886                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
4887                         break;
4888                 if (i == 2) {
4889                         e_err("The NVM Checksum Is Not Valid\n");
4890                         err = -EIO;
4891                         goto err_eeprom;
4892                 }
4893         }
4894
4895         e1000_eeprom_checks(adapter);
4896
4897         /* copy the MAC address out of the NVM */
4898         if (e1000e_read_mac_addr(&adapter->hw))
4899                 e_err("NVM Read Error while reading MAC address\n");
4900
4901         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
4902         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
4903
4904         if (!is_valid_ether_addr(netdev->perm_addr)) {
4905                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
4906                 err = -EIO;
4907                 goto err_eeprom;
4908         }
4909
4910         init_timer(&adapter->watchdog_timer);
4911         adapter->watchdog_timer.function = &e1000_watchdog;
4912         adapter->watchdog_timer.data = (unsigned long) adapter;
4913
4914         init_timer(&adapter->phy_info_timer);
4915         adapter->phy_info_timer.function = &e1000_update_phy_info;
4916         adapter->phy_info_timer.data = (unsigned long) adapter;
4917
4918         INIT_WORK(&adapter->reset_task, e1000_reset_task);
4919         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
4920         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
4921         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
4922
4923         /* Initialize link parameters. User can change them with ethtool */
4924         adapter->hw.mac.autoneg = 1;
4925         adapter->fc_autoneg = 1;
4926         adapter->hw.fc.original_type = e1000_fc_default;
4927         adapter->hw.fc.type = e1000_fc_default;
4928         adapter->hw.phy.autoneg_advertised = 0x2f;
4929
4930         /* ring size defaults */
4931         adapter->rx_ring->count = 256;
4932         adapter->tx_ring->count = 256;
4933
4934         /*
4935          * Initial Wake on LAN setting - If APM wake is enabled in
4936          * the EEPROM, enable the ACPI Magic Packet filter
4937          */
4938         if (adapter->flags & FLAG_APME_IN_WUC) {
4939                 /* APME bit in EEPROM is mapped to WUC.APME */
4940                 eeprom_data = er32(WUC);
4941                 eeprom_apme_mask = E1000_WUC_APME;
4942         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
4943                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
4944                     (adapter->hw.bus.func == 1))
4945                         e1000_read_nvm(&adapter->hw,
4946                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
4947                 else
4948                         e1000_read_nvm(&adapter->hw,
4949                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
4950         }
4951
4952         /* fetch WoL from EEPROM */
4953         if (eeprom_data & eeprom_apme_mask)
4954                 adapter->eeprom_wol |= E1000_WUFC_MAG;
4955
4956         /*
4957          * now that we have the eeprom settings, apply the special cases
4958          * where the eeprom may be wrong or the board simply won't support
4959          * wake on lan on a particular port
4960          */
4961         if (!(adapter->flags & FLAG_HAS_WOL))
4962                 adapter->eeprom_wol = 0;
4963
4964         /* initialize the wol settings based on the eeprom settings */
4965         adapter->wol = adapter->eeprom_wol;
4966
4967         /* reset the hardware with the new settings */
4968         e1000e_reset(adapter);
4969
4970         /*
4971          * If the controller has AMT, do not set DRV_LOAD until the interface
4972          * is up.  For all other cases, let the f/w know that the h/w is now
4973          * under the control of the driver.
4974          */
4975         if (!(adapter->flags & FLAG_HAS_AMT))
4976                 e1000_get_hw_control(adapter);
4977
4978         /* tell the stack to leave us alone until e1000_open() is called */
4979         netif_carrier_off(netdev);
4980         netif_tx_stop_all_queues(netdev);
4981
4982         strcpy(netdev->name, "eth%d");
4983         err = register_netdev(netdev);
4984         if (err)
4985                 goto err_register;
4986
4987         e1000_print_device_info(adapter);
4988
4989         return 0;
4990
4991 err_register:
4992         if (!(adapter->flags & FLAG_HAS_AMT))
4993                 e1000_release_hw_control(adapter);
4994 err_eeprom:
4995         if (!e1000_check_reset_block(&adapter->hw))
4996                 e1000_phy_hw_reset(&adapter->hw);
4997 err_hw_init:
4998
4999         kfree(adapter->tx_ring);
5000         kfree(adapter->rx_ring);
5001 err_sw_init:
5002         if (adapter->hw.flash_address)
5003                 iounmap(adapter->hw.flash_address);
5004 err_flashmap:
5005         iounmap(adapter->hw.hw_addr);
5006 err_ioremap:
5007         free_netdev(netdev);
5008 err_alloc_etherdev:
5009         pci_release_selected_regions(pdev,
5010                                      pci_select_bars(pdev, IORESOURCE_MEM));
5011 err_pci_reg:
5012 err_dma:
5013         pci_disable_device(pdev);
5014         return err;
5015 }
5016
5017 /**
5018  * e1000_remove - Device Removal Routine
5019  * @pdev: PCI device information struct
5020  *
5021  * e1000_remove is called by the PCI subsystem to alert the driver
5022  * that it should release a PCI device.  The could be caused by a
5023  * Hot-Plug event, or because the driver is going to be removed from
5024  * memory.
5025  **/
5026 static void __devexit e1000_remove(struct pci_dev *pdev)
5027 {
5028         struct net_device *netdev = pci_get_drvdata(pdev);
5029         struct e1000_adapter *adapter = netdev_priv(netdev);
5030
5031         /*
5032          * flush_scheduled work may reschedule our watchdog task, so
5033          * explicitly disable watchdog tasks from being rescheduled
5034          */
5035         set_bit(__E1000_DOWN, &adapter->state);
5036         del_timer_sync(&adapter->watchdog_timer);
5037         del_timer_sync(&adapter->phy_info_timer);
5038
5039         flush_scheduled_work();
5040
5041         /*
5042          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5043          * would have already happened in close and is redundant.
5044          */
5045         e1000_release_hw_control(adapter);
5046
5047         unregister_netdev(netdev);
5048
5049         if (!e1000_check_reset_block(&adapter->hw))
5050                 e1000_phy_hw_reset(&adapter->hw);
5051
5052         e1000e_reset_interrupt_capability(adapter);
5053         kfree(adapter->tx_ring);
5054         kfree(adapter->rx_ring);
5055
5056         iounmap(adapter->hw.hw_addr);
5057         if (adapter->hw.flash_address)
5058                 iounmap(adapter->hw.flash_address);
5059         pci_release_selected_regions(pdev,
5060                                      pci_select_bars(pdev, IORESOURCE_MEM));
5061
5062         free_netdev(netdev);
5063
5064         pci_disable_device(pdev);
5065 }
5066
5067 /* PCI Error Recovery (ERS) */
5068 static struct pci_error_handlers e1000_err_handler = {
5069         .error_detected = e1000_io_error_detected,
5070         .slot_reset = e1000_io_slot_reset,
5071         .resume = e1000_io_resume,
5072 };
5073
5074 static struct pci_device_id e1000_pci_tbl[] = {
5075         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5076         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5077         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5078         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5079         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5080         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5081         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5082         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5083         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5084
5085         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5086         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5087         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5088         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5089
5090         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5091         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5092         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5093
5094         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5095
5096         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5097           board_80003es2lan },
5098         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5099           board_80003es2lan },
5100         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5101           board_80003es2lan },
5102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5103           board_80003es2lan },
5104
5105         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5106         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5108         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5109         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5110         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5111         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5112
5113         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5114         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5115         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5116         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5117         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5118         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5120         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5122
5123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5124         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5126
5127         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5128         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5129
5130         { }     /* terminate list */
5131 };
5132 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5133
5134 /* PCI Device API Driver */
5135 static struct pci_driver e1000_driver = {
5136         .name     = e1000e_driver_name,
5137         .id_table = e1000_pci_tbl,
5138         .probe    = e1000_probe,
5139         .remove   = __devexit_p(e1000_remove),
5140 #ifdef CONFIG_PM
5141         /* Power Management Hooks */
5142         .suspend  = e1000_suspend,
5143         .resume   = e1000_resume,
5144 #endif
5145         .shutdown = e1000_shutdown,
5146         .err_handler = &e1000_err_handler
5147 };
5148
5149 /**
5150  * e1000_init_module - Driver Registration Routine
5151  *
5152  * e1000_init_module is the first routine called when the driver is
5153  * loaded. All it does is register with the PCI subsystem.
5154  **/
5155 static int __init e1000_init_module(void)
5156 {
5157         int ret;
5158         printk(KERN_INFO "%s: Intel(R) PRO/1000 Network Driver - %s\n",
5159                e1000e_driver_name, e1000e_driver_version);
5160         printk(KERN_INFO "%s: Copyright (c) 1999-2008 Intel Corporation.\n",
5161                e1000e_driver_name);
5162         ret = pci_register_driver(&e1000_driver);
5163         pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name,
5164                                PM_QOS_DEFAULT_VALUE);
5165
5166         return ret;
5167 }
5168 module_init(e1000_init_module);
5169
5170 /**
5171  * e1000_exit_module - Driver Exit Cleanup Routine
5172  *
5173  * e1000_exit_module is called just before the driver is removed
5174  * from memory.
5175  **/
5176 static void __exit e1000_exit_module(void)
5177 {
5178         pci_unregister_driver(&e1000_driver);
5179         pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY, e1000e_driver_name);
5180 }
5181 module_exit(e1000_exit_module);
5182
5183
5184 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5185 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5186 MODULE_LICENSE("GPL");
5187 MODULE_VERSION(DRV_VERSION);
5188
5189 /* e1000_main.c */